EchoSwift-700M (LiteRT-LM Android Packages)
Official mobile-quantized Google LiteRT (.litertlm) binaries for adityavardhansharma/EchoSwift-700M, purpose-built for on-device deployment in EchoFlow.
📱 Available Quantized Mobile Packages
Directly import these into EchoFlow:
| Package File | Context Window | Quantization | Size | Target Hardware | Direct Download |
|---|---|---|---|---|---|
EchoSwift-700M_int8.litertlm |
4,096 tokens | INT8 Dynamic (dynamic_wi8_afp32) |
770.8 MB | Modern Android Phones (Snapdragon NPU, Dimensity APU) | Download 4K INT8 |
EchoSwift-700M_int4.litertlm |
4,096 tokens | INT4 Blockwise-32 (wi4b32 OCTAV) |
458.5 MB | Ultra-lightweight / RAM-constrained phones | Download 4K INT4 |
EchoSwift-700M_int8_8k_ekv8192.litertlm |
8,192 tokens | INT8 Dynamic (dynamic_wi8_afp32) |
770.8 MB | Extended Context (Long chats & complex tools) | Download 8K INT8 |
EchoSwift-700M_int4_8k_ekv8192.litertlm |
8,192 tokens | INT4 Blockwise-32 (wi4b32 OCTAV) |
460.0 MB | Extended Context on budget phones | Download 8K INT4 |
⚡ Highlights
- 100.0% EchoFlow Schema Pass Rate: Flawless adherence to single-line search triggers (
search: query) and single-file HTML mobile artifacts. - Compiled with FlatBuffer & 11 Prefill Buckets: Optimized across prefill lengths (1 to 1024) and 4096 KV-cache length for Qualcomm Hexagon/Adreno and MediaTek APUs.
Base model: adityavardhansharma/EchoSwift-700M
Application website: https://echoflow.adityavs.space