lfm25-soap-gguf

LiquidAI LFM2.5 1.2B fine-tuned for clinical SOAP notes, exported as mixed-precision GGUF for on-device use (llama.cpp / llama.rn, CPU).

Files

File Size Role
lfm25-soap-IQ3_M.gguf 599 MB Ship model for 8 GB Android phones (IQ quant, q8_0 embeddings/lm_head, SOAP imatrix)
lfm25-soap-Q4_K_M.gguf 731 MB Fallback if IQ3_M is unavailable in the runtime
lfm25-soap-Q6_K.gguf 963 MB Reference quality on 12 GB+ devices
lfm25_edge_android_config.json 1 KB Runtime config (context, threads, CPU-only, stop token)
soap_prompt.ts 6 KB Prompt builder buildLfm25SoapPrompt and parsers (parseFields, splitCompletion)
INTEGRATION.md 1 KB Integration notes

Prompt format

Chat template <|im_start|>{role}\n{content}, stop token <|im_end|>. Do not use Llama <|eot_id|> tokens with this model.

Runtime settings

  • CPU only (n_gpu_layers: 0), 4 threads, mmap on
  • Context 3072, max generated tokens 1280
  • Temperature 0.0, top_p 1.0
  • Dialogue clipped to 2800 characters

Not a diagnostic device. Outputs must be reviewed by a clinician.

Downloads last month
139
GGUF
Model size
1B params
Architecture
lfm2
Hardware compatibility
Log In to add your hardware

3-bit

4-bit

6-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support