pre-release (11)

added_tokens.json

@@ -1,8 +1,60 @@
 {
+  "</caption>": 49192,
+  "</chart>": 49248,
+  "</checkbox_selected>": 49211,
+  "</checkbox_unselected>": 49213,
+  "</doctag>": 49230,
+  "</footnote>": 49195,
+  "</form>": 49215,
+  "</formula>": 49197,
+  "</group>": 49228,
+  "</key_": 49243,
+  "</key_value_region>": 49217,
+  "</list_item>": 49199,
+  "</ordered_list>": 49224,
+  "</otsl>": 49209,
+  "</page_footer>": 49201,
+  "</page_header>": 49203,
+  "</paragraph>": 49220,
+  "</picture>": 49205,
+  "</reference>": 49222,
+  "</section_header_level_": 49207,
+  "</smiles>": 49251,
+  "</unordered_list>": 49226,
+  "</value_": 49245,
+  "<caption>": 49191,
+  "<chart>": 49247,
+  "<checkbox_selected>": 49210,
+  "<checkbox_unselected>": 49212,
+  "<ched>": 49239,
+  "<doctag>": 49229,
+  "<ecel>": 49234,
   "<end_of_utterance>": 49279,
   "<fake_token_around_image>": 49189,
+  "<fcel>": 49233,
+  "<footnote>": 49193,
+  "<form>": 49214,
+  "<formula>": 49196,
   "<global-img>": 49152,
+  "<group>": 49227,
   "<image>": 49190,
+  "<key_": 49242,
+  "<key_value_region>": 49216,
+  "<lcel>": 49235,
+  "<link_": 49246,
+  "<list_item>": 49198,
+  "<loc_": 49218,
+  "<nl>": 49238,
+  "<ordered_list>": 49223,
+  "<otsl>": 49208,
+  "<page_": 49231,
+  "<page_break>": 49249,
+  "<page_footer>": 49200,
+  "<page_header>": 49202,
+  "<paragraph>": 49219,
+  "<picture>": 49204,
+  "<reference>": 49221,
+  "<rhed>": 49240,
   "<row_1_col_1>": 49153,
   "<row_1_col_2>": 49154,
   "<row_1_col_3>": 49155,
@@ -39,63 +91,15 @@
   "<row_6_col_4>": 49186,
   "<row_6_col_5>": 49187,
   "<row_6_col_6>": 49188,
-  "<|reserved_special_token_0|>": 49191,
-  "<|reserved_special_token_10|>": 49201,
-  "<|reserved_special_token_11|>": 49202,
-  "<|reserved_special_token_12|>": 49203,
-  "<|reserved_special_token_13|>": 49204,
-  "<|reserved_special_token_14|>": 49205,
-  "<|reserved_special_token_15|>": 49206,
-  "<|reserved_special_token_16|>": 49207,
-  "<|reserved_special_token_17|>": 49208,
-  "<|reserved_special_token_18|>": 49209,
-  "<|reserved_special_token_19|>": 49210,
-  "<|reserved_special_token_1|>": 49192,
-  "<|reserved_special_token_20|>": 49211,
-  "<|reserved_special_token_21|>": 49212,
-  "<|reserved_special_token_22|>": 49213,
-  "<|reserved_special_token_23|>": 49214,
-  "<|reserved_special_token_24|>": 49215,
-  "<|reserved_special_token_25|>": 49216,
-  "<|reserved_special_token_26|>": 49217,
-  "<|reserved_special_token_27|>": 49218,
-  "<|reserved_special_token_28|>": 49219,
-  "<|reserved_special_token_29|>": 49220,
-  "<|reserved_special_token_2|>": 49193,
-  "<|reserved_special_token_30|>": 49221,
-  "<|reserved_special_token_31|>": 49222,
-  "<|reserved_special_token_32|>": 49223,
-  "<|reserved_special_token_33|>": 49224,
-  "<|reserved_special_token_34|>": 49225,
-  "<|reserved_special_token_35|>": 49226,
-  "<|reserved_special_token_36|>": 49227,
-  "<|reserved_special_token_37|>": 49228,
-  "<|reserved_special_token_38|>": 49229,
-  "<|reserved_special_token_39|>": 49230,
+  "<section_header_level_": 49206,
+  "<smiles>": 49250,
+  "<text_break>": 49232,
+  "<ucel>": 49236,
+  "<unordered_list>": 49225,
+  "<value_": 49244,
+  "<xcel>": 49237,
   "<|reserved_special_token_3|>": 49194,
-  "<|reserved_special_token_40|>": 49231,
-  "<|reserved_special_token_41|>": 49232,
-  "<|reserved_special_token_42|>": 49233,
-  "<|reserved_special_token_43|>": 49234,
-  "<|reserved_special_token_44|>": 49235,
-  "<|reserved_special_token_45|>": 49236,
-  "<|reserved_special_token_46|>": 49237,
-  "<|reserved_special_token_47|>": 49238,
-  "<|reserved_special_token_48|>": 49239,
-  "<|reserved_special_token_49|>": 49240,
-  "<|reserved_special_token_4|>": 49195,
   "<|reserved_special_token_50|>": 49241,
-  "<|reserved_special_token_51|>": 49242,
-  "<|reserved_special_token_52|>": 49243,
-  "<|reserved_special_token_53|>": 49244,
-  "<|reserved_special_token_54|>": 49245,
-  "<|reserved_special_token_55|>": 49246,
-  "<|reserved_special_token_56|>": 49247,
-  "<|reserved_special_token_57|>": 49248,
-  "<|reserved_special_token_58|>": 49249,
-  "<|reserved_special_token_59|>": 49250,
-  "<|reserved_special_token_5|>": 49196,
-  "<|reserved_special_token_60|>": 49251,
   "<|reserved_special_token_61|>": 49252,
   "<|reserved_special_token_62|>": 49253,
   "<|reserved_special_token_63|>": 49254,
@@ -105,7 +109,6 @@
   "<|reserved_special_token_67|>": 49258,
   "<|reserved_special_token_68|>": 49259,
   "<|reserved_special_token_69|>": 49260,
-  "<|reserved_special_token_6|>": 49197,
   "<|reserved_special_token_70|>": 49261,
   "<|reserved_special_token_71|>": 49262,
   "<|reserved_special_token_72|>": 49263,
@@ -116,7 +119,6 @@
   "<|reserved_special_token_77|>": 49268,
   "<|reserved_special_token_78|>": 49269,
   "<|reserved_special_token_79|>": 49270,
-  "<|reserved_special_token_7|>": 49198,
   "<|reserved_special_token_80|>": 49271,
   "<|reserved_special_token_81|>": 49272,
   "<|reserved_special_token_82|>": 49273,
@@ -124,7 +126,5 @@
   "<|reserved_special_token_84|>": 49275,
   "<|reserved_special_token_85|>": 49276,
   "<|reserved_special_token_86|>": 49277,
-  "<|reserved_special_token_87|>": 49278,
-  "<|reserved_special_token_8|>": 49199,
-  "<|reserved_special_token_9|>": 49200
+  "<|reserved_special_token_87|>": 49278
 }

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/dccstor/ahn_cdip/checkpoints/SmolDocling_250M_1.3_stg_6/SmolDocling-256M-preview",
+  "_name_or_path": "/data1/checkpoints/SmolDocling_250M_DT_ST10/checkpoint-1758/",
   "architectures": [
     "Idefics3ForConditionalGeneration"
   ],
@@ -115,7 +115,7 @@
   },
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
-  "transformers_version": "4.48.3",
+  "transformers_version": "4.50.0.dev0",
   "use_cache": true,
   "vision_config": {
     "hidden_size": 768,

generation_config.json

@@ -3,5 +3,5 @@
   "bos_token_id": 0,
   "eos_token_id": 49279,
   "pad_token_id": 2,
-  "transformers_version": "4.48.3"
+  "transformers_version": "4.50.0.dev0"
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa304312ea5d92e0aadf70ff5e3f330ddd289ec9b93dd8cb1cdb438494e17709
+oid sha256:cdcdf5d823c5684029c7d8e52177cf10f9034b3aba6577549cfb1a9ce36ad0a2
 size 513028808

tokenizer.json

@@ -1,7 +1,19 @@
 {
   "version": "1.0",
-  "truncation": null,
-  "padding": null,
+  "truncation": {
+    "direction": "Right",
+    "max_length": 8192,
+    "strategy": "LongestFirst",
+    "stride": 0
+  },
+  "padding": {
+    "strategy": "BatchLongest",
+    "direction": "Right",
+    "pad_to_multiple_of": null,
+    "pad_id": 2,
+    "pad_type_id": 0,
+    "pad_token": "<|im_end|>"
+  },
   "added_tokens": [
     {
       "id": 0,
@@ -509,7 +521,7 @@
     },
     {
       "id": 49191,
-      "content": "<|reserved_special_token_0|>",
+      "content": "<caption>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -518,7 +530,7 @@
     },
     {
       "id": 49192,
-      "content": "<|reserved_special_token_1|>",
+      "content": "</caption>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -527,7 +539,7 @@
     },
     {
       "id": 49193,
-      "content": "<|reserved_special_token_2|>",
+      "content": "<footnote>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -545,7 +557,7 @@
     },
     {
       "id": 49195,
-      "content": "<|reserved_special_token_4|>",
+      "content": "</footnote>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -554,7 +566,7 @@
     },
     {
       "id": 49196,
-      "content": "<|reserved_special_token_5|>",
+      "content": "<formula>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -563,7 +575,7 @@
     },
     {
       "id": 49197,
-      "content": "<|reserved_special_token_6|>",
+      "content": "</formula>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -572,7 +584,7 @@
     },
     {
       "id": 49198,
-      "content": "<|reserved_special_token_7|>",
+      "content": "<list_item>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -581,7 +593,7 @@
     },
     {
       "id": 49199,
-      "content": "<|reserved_special_token_8|>",
+      "content": "</list_item>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -590,7 +602,7 @@
     },
     {
       "id": 49200,
-      "content": "<|reserved_special_token_9|>",
+      "content": "<page_footer>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -599,7 +611,7 @@
     },
     {
       "id": 49201,
-      "content": "<|reserved_special_token_10|>",
+      "content": "</page_footer>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -608,7 +620,7 @@
     },
     {
       "id": 49202,
-      "content": "<|reserved_special_token_11|>",
+      "content": "<page_header>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -617,7 +629,7 @@
     },
     {
       "id": 49203,
-      "content": "<|reserved_special_token_12|>",
+      "content": "</page_header>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -626,7 +638,7 @@
     },
     {
       "id": 49204,
-      "content": "<|reserved_special_token_13|>",
+      "content": "<picture>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -635,7 +647,7 @@
     },
     {
       "id": 49205,
-      "content": "<|reserved_special_token_14|>",
+      "content": "</picture>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -644,7 +656,7 @@
     },
     {
       "id": 49206,
-      "content": "<|reserved_special_token_15|>",
+      "content": "<section_header_level_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -653,7 +665,7 @@
     },
     {
       "id": 49207,
-      "content": "<|reserved_special_token_16|>",
+      "content": "</section_header_level_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -662,7 +674,7 @@
     },
     {
       "id": 49208,
-      "content": "<|reserved_special_token_17|>",
+      "content": "<otsl>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -671,7 +683,7 @@
     },
     {
       "id": 49209,
-      "content": "<|reserved_special_token_18|>",
+      "content": "</otsl>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -680,7 +692,7 @@
     },
     {
       "id": 49210,
-      "content": "<|reserved_special_token_19|>",
+      "content": "<checkbox_selected>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -689,7 +701,7 @@
     },
     {
       "id": 49211,
-      "content": "<|reserved_special_token_20|>",
+      "content": "</checkbox_selected>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -698,7 +710,7 @@
     },
     {
       "id": 49212,
-      "content": "<|reserved_special_token_21|>",
+      "content": "<checkbox_unselected>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -707,7 +719,7 @@
     },
     {
       "id": 49213,
-      "content": "<|reserved_special_token_22|>",
+      "content": "</checkbox_unselected>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -716,7 +728,7 @@
     },
     {
       "id": 49214,
-      "content": "<|reserved_special_token_23|>",
+      "content": "<form>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -725,7 +737,7 @@
     },
     {
       "id": 49215,
-      "content": "<|reserved_special_token_24|>",
+      "content": "</form>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -734,7 +746,7 @@
     },
     {
       "id": 49216,
-      "content": "<|reserved_special_token_25|>",
+      "content": "<key_value_region>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -743,7 +755,7 @@
     },
     {
       "id": 49217,
-      "content": "<|reserved_special_token_26|>",
+      "content": "</key_value_region>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -752,7 +764,7 @@
     },
     {
       "id": 49218,
-      "content": "<|reserved_special_token_27|>",
+      "content": "<loc_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -761,7 +773,7 @@
     },
     {
       "id": 49219,
-      "content": "<|reserved_special_token_28|>",
+      "content": "<paragraph>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -770,7 +782,7 @@
     },
     {
       "id": 49220,
-      "content": "<|reserved_special_token_29|>",
+      "content": "</paragraph>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -779,7 +791,7 @@
     },
     {
       "id": 49221,
-      "content": "<|reserved_special_token_30|>",
+      "content": "<reference>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -788,7 +800,7 @@
     },
     {
       "id": 49222,
-      "content": "<|reserved_special_token_31|>",
+      "content": "</reference>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -797,7 +809,7 @@
     },
     {
       "id": 49223,
-      "content": "<|reserved_special_token_32|>",
+      "content": "<ordered_list>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -806,7 +818,7 @@
     },
     {
       "id": 49224,
-      "content": "<|reserved_special_token_33|>",
+      "content": "</ordered_list>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -815,7 +827,7 @@
     },
     {
       "id": 49225,
-      "content": "<|reserved_special_token_34|>",
+      "content": "<unordered_list>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -824,7 +836,7 @@
     },
     {
       "id": 49226,
-      "content": "<|reserved_special_token_35|>",
+      "content": "</unordered_list>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -833,7 +845,7 @@
     },
     {
       "id": 49227,
-      "content": "<|reserved_special_token_36|>",
+      "content": "<group>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -842,7 +854,7 @@
     },
     {
       "id": 49228,
-      "content": "<|reserved_special_token_37|>",
+      "content": "</group>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -851,7 +863,7 @@
     },
     {
       "id": 49229,
-      "content": "<|reserved_special_token_38|>",
+      "content": "<doctag>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -860,7 +872,7 @@
     },
     {
       "id": 49230,
-      "content": "<|reserved_special_token_39|>",
+      "content": "</doctag>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -869,7 +881,7 @@
     },
     {
       "id": 49231,
-      "content": "<|reserved_special_token_40|>",
+      "content": "<page_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -878,7 +890,7 @@
     },
     {
       "id": 49232,
-      "content": "<|reserved_special_token_41|>",
+      "content": "<text_break>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -887,7 +899,7 @@
     },
     {
       "id": 49233,
-      "content": "<|reserved_special_token_42|>",
+      "content": "<fcel>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -896,7 +908,7 @@
     },
     {
       "id": 49234,
-      "content": "<|reserved_special_token_43|>",
+      "content": "<ecel>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -905,7 +917,7 @@
     },
     {
       "id": 49235,
-      "content": "<|reserved_special_token_44|>",
+      "content": "<lcel>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -914,7 +926,7 @@
     },
     {
       "id": 49236,
-      "content": "<|reserved_special_token_45|>",
+      "content": "<ucel>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -923,7 +935,7 @@
     },
     {
       "id": 49237,
-      "content": "<|reserved_special_token_46|>",
+      "content": "<xcel>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -932,7 +944,7 @@
     },
     {
       "id": 49238,
-      "content": "<|reserved_special_token_47|>",
+      "content": "<nl>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -941,7 +953,7 @@
     },
     {
       "id": 49239,
-      "content": "<|reserved_special_token_48|>",
+      "content": "<ched>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -950,7 +962,7 @@
     },
     {
       "id": 49240,
-      "content": "<|reserved_special_token_49|>",
+      "content": "<rhed>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -968,7 +980,7 @@
     },
     {
       "id": 49242,
-      "content": "<|reserved_special_token_51|>",
+      "content": "<key_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -977,7 +989,7 @@
     },
     {
       "id": 49243,
-      "content": "<|reserved_special_token_52|>",
+      "content": "</key_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -986,7 +998,7 @@
     },
     {
       "id": 49244,
-      "content": "<|reserved_special_token_53|>",
+      "content": "<value_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -995,7 +1007,7 @@
     },
     {
       "id": 49245,
-      "content": "<|reserved_special_token_54|>",
+      "content": "</value_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -1004,7 +1016,7 @@
     },
     {
       "id": 49246,
-      "content": "<|reserved_special_token_55|>",
+      "content": "<link_",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -1013,7 +1025,7 @@
     },
     {
       "id": 49247,
-      "content": "<|reserved_special_token_56|>",
+      "content": "<chart>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -1022,7 +1034,7 @@
     },
     {
       "id": 49248,
-      "content": "<|reserved_special_token_57|>",
+      "content": "</chart>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -1031,7 +1043,7 @@
     },
     {
       "id": 49249,
-      "content": "<|reserved_special_token_58|>",
+      "content": "<page_break>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -1040,7 +1052,7 @@
     },
     {
       "id": 49250,
-      "content": "<|reserved_special_token_59|>",
+      "content": "<smiles>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,
@@ -1049,7 +1061,7 @@
     },
     {
       "id": 49251,
-      "content": "<|reserved_special_token_60|>",
+      "content": "</smiles>",
       "single_word": false,
       "lstrip": false,
       "rstrip": false,

tokenizer_config.json

@@ -450,7 +450,7 @@
       "special": true
     },
     "49191": {
-      "content": "<|reserved_special_token_0|>",
+      "content": "<caption>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -458,7 +458,7 @@
       "special": true
     },
     "49192": {
-      "content": "<|reserved_special_token_1|>",
+      "content": "</caption>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -466,7 +466,7 @@
       "special": true
     },
     "49193": {
-      "content": "<|reserved_special_token_2|>",
+      "content": "<footnote>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -482,7 +482,7 @@
       "special": true
     },
     "49195": {
-      "content": "<|reserved_special_token_4|>",
+      "content": "</footnote>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -490,7 +490,7 @@
       "special": true
     },
     "49196": {
-      "content": "<|reserved_special_token_5|>",
+      "content": "<formula>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -498,7 +498,7 @@
       "special": true
     },
     "49197": {
-      "content": "<|reserved_special_token_6|>",
+      "content": "</formula>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -506,7 +506,7 @@
       "special": true
     },
     "49198": {
-      "content": "<|reserved_special_token_7|>",
+      "content": "<list_item>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -514,7 +514,7 @@
       "special": true
     },
     "49199": {
-      "content": "<|reserved_special_token_8|>",
+      "content": "</list_item>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -522,7 +522,7 @@
       "special": true
     },
     "49200": {
-      "content": "<|reserved_special_token_9|>",
+      "content": "<page_footer>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -530,7 +530,7 @@
       "special": true
     },
     "49201": {
-      "content": "<|reserved_special_token_10|>",
+      "content": "</page_footer>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -538,7 +538,7 @@
       "special": true
     },
     "49202": {
-      "content": "<|reserved_special_token_11|>",
+      "content": "<page_header>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -546,7 +546,7 @@
       "special": true
     },
     "49203": {
-      "content": "<|reserved_special_token_12|>",
+      "content": "</page_header>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -554,7 +554,7 @@
       "special": true
     },
     "49204": {
-      "content": "<|reserved_special_token_13|>",
+      "content": "<picture>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -562,7 +562,7 @@
       "special": true
     },
     "49205": {
-      "content": "<|reserved_special_token_14|>",
+      "content": "</picture>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -570,7 +570,7 @@
       "special": true
     },
     "49206": {
-      "content": "<|reserved_special_token_15|>",
+      "content": "<section_header_level_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -578,7 +578,7 @@
       "special": true
     },
     "49207": {
-      "content": "<|reserved_special_token_16|>",
+      "content": "</section_header_level_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -586,7 +586,7 @@
       "special": true
     },
     "49208": {
-      "content": "<|reserved_special_token_17|>",
+      "content": "<otsl>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -594,7 +594,7 @@
       "special": true
     },
     "49209": {
-      "content": "<|reserved_special_token_18|>",
+      "content": "</otsl>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -602,7 +602,7 @@
       "special": true
     },
     "49210": {
-      "content": "<|reserved_special_token_19|>",
+      "content": "<checkbox_selected>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -610,7 +610,7 @@
       "special": true
     },
     "49211": {
-      "content": "<|reserved_special_token_20|>",
+      "content": "</checkbox_selected>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -618,7 +618,7 @@
       "special": true
     },
     "49212": {
-      "content": "<|reserved_special_token_21|>",
+      "content": "<checkbox_unselected>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -626,7 +626,7 @@
       "special": true
     },
     "49213": {
-      "content": "<|reserved_special_token_22|>",
+      "content": "</checkbox_unselected>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -634,7 +634,7 @@
       "special": true
     },
     "49214": {
-      "content": "<|reserved_special_token_23|>",
+      "content": "<form>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -642,7 +642,7 @@
       "special": true
     },
     "49215": {
-      "content": "<|reserved_special_token_24|>",
+      "content": "</form>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -650,7 +650,7 @@
       "special": true
     },
     "49216": {
-      "content": "<|reserved_special_token_25|>",
+      "content": "<key_value_region>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -658,7 +658,7 @@
       "special": true
     },
     "49217": {
-      "content": "<|reserved_special_token_26|>",
+      "content": "</key_value_region>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -666,7 +666,7 @@
       "special": true
     },
     "49218": {
-      "content": "<|reserved_special_token_27|>",
+      "content": "<loc_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -674,7 +674,7 @@
       "special": true
     },
     "49219": {
-      "content": "<|reserved_special_token_28|>",
+      "content": "<paragraph>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -682,7 +682,7 @@
       "special": true
     },
     "49220": {
-      "content": "<|reserved_special_token_29|>",
+      "content": "</paragraph>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -690,7 +690,7 @@
       "special": true
     },
     "49221": {
-      "content": "<|reserved_special_token_30|>",
+      "content": "<reference>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -698,7 +698,7 @@
       "special": true
     },
     "49222": {
-      "content": "<|reserved_special_token_31|>",
+      "content": "</reference>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -706,7 +706,7 @@
       "special": true
     },
     "49223": {
-      "content": "<|reserved_special_token_32|>",
+      "content": "<ordered_list>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -714,7 +714,7 @@
       "special": true
     },
     "49224": {
-      "content": "<|reserved_special_token_33|>",
+      "content": "</ordered_list>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -722,7 +722,7 @@
       "special": true
     },
     "49225": {
-      "content": "<|reserved_special_token_34|>",
+      "content": "<unordered_list>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -730,7 +730,7 @@
       "special": true
     },
     "49226": {
-      "content": "<|reserved_special_token_35|>",
+      "content": "</unordered_list>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -738,7 +738,7 @@
       "special": true
     },
     "49227": {
-      "content": "<|reserved_special_token_36|>",
+      "content": "<group>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -746,7 +746,7 @@
       "special": true
     },
     "49228": {
-      "content": "<|reserved_special_token_37|>",
+      "content": "</group>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -754,7 +754,7 @@
       "special": true
     },
     "49229": {
-      "content": "<|reserved_special_token_38|>",
+      "content": "<doctag>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -762,7 +762,7 @@
       "special": true
     },
     "49230": {
-      "content": "<|reserved_special_token_39|>",
+      "content": "</doctag>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -770,7 +770,7 @@
       "special": true
     },
     "49231": {
-      "content": "<|reserved_special_token_40|>",
+      "content": "<page_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -778,7 +778,7 @@
       "special": true
     },
     "49232": {
-      "content": "<|reserved_special_token_41|>",
+      "content": "<text_break>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -786,7 +786,7 @@
       "special": true
     },
     "49233": {
-      "content": "<|reserved_special_token_42|>",
+      "content": "<fcel>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -794,7 +794,7 @@
       "special": true
     },
     "49234": {
-      "content": "<|reserved_special_token_43|>",
+      "content": "<ecel>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -802,7 +802,7 @@
       "special": true
     },
     "49235": {
-      "content": "<|reserved_special_token_44|>",
+      "content": "<lcel>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -810,7 +810,7 @@
       "special": true
     },
     "49236": {
-      "content": "<|reserved_special_token_45|>",
+      "content": "<ucel>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -818,7 +818,7 @@
       "special": true
     },
     "49237": {
-      "content": "<|reserved_special_token_46|>",
+      "content": "<xcel>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -826,7 +826,7 @@
       "special": true
     },
     "49238": {
-      "content": "<|reserved_special_token_47|>",
+      "content": "<nl>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -834,7 +834,7 @@
       "special": true
     },
     "49239": {
-      "content": "<|reserved_special_token_48|>",
+      "content": "<ched>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -842,7 +842,7 @@
       "special": true
     },
     "49240": {
-      "content": "<|reserved_special_token_49|>",
+      "content": "<rhed>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -858,7 +858,7 @@
       "special": true
     },
     "49242": {
-      "content": "<|reserved_special_token_51|>",
+      "content": "<key_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -866,7 +866,7 @@
       "special": true
     },
     "49243": {
-      "content": "<|reserved_special_token_52|>",
+      "content": "</key_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -874,7 +874,7 @@
       "special": true
     },
     "49244": {
-      "content": "<|reserved_special_token_53|>",
+      "content": "<value_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -882,7 +882,7 @@
       "special": true
     },
     "49245": {
-      "content": "<|reserved_special_token_54|>",
+      "content": "</value_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -890,7 +890,7 @@
       "special": true
     },
     "49246": {
-      "content": "<|reserved_special_token_55|>",
+      "content": "<link_",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -898,7 +898,7 @@
       "special": true
     },
     "49247": {
-      "content": "<|reserved_special_token_56|>",
+      "content": "<chart>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -906,7 +906,7 @@
       "special": true
     },
     "49248": {
-      "content": "<|reserved_special_token_57|>",
+      "content": "</chart>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -914,7 +914,7 @@
       "special": true
     },
     "49249": {
-      "content": "<|reserved_special_token_58|>",
+      "content": "<page_break>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -922,7 +922,7 @@
       "special": true
     },
     "49250": {
-      "content": "<|reserved_special_token_59|>",
+      "content": "<smiles>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -930,7 +930,7 @@
       "special": true
     },
     "49251": {
-      "content": "<|reserved_special_token_60|>",
+      "content": "</smiles>",
       "lstrip": false,
       "normalized": false,
       "rstrip": false,
@@ -1173,11 +1173,17 @@
   "eos_token": "<|im_end|>",
   "extra_special_tokens": {},
   "legacy": false,
+  "max_length": 8192,
   "model_max_length": 8192,
+  "pad_to_multiple_of": null,
   "pad_token": "<|im_end|>",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
   "processor_class": "Idefics3Processor",
+  "stride": 0,
   "tokenizer_class": "GPT2Tokenizer",
-  "truncation_side": "left",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
   "unk_token": "<|endoftext|>",
   "vocab_size": 49152
 }

zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+#   python zero_to_fp32.py . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device, weights_only=False)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+    total_files = len(files)
+    state_dicts = []
+    for f in tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)