LFM2.5-350M — ExecuTorch XNNPACK 8da4w
lfm2_5_350m_xnnpack_8da4w.pte (252.8 MB)
- Source: LiquidAI/LFM2.5-350M (hybrid conv/attention, 16 layers)
- License: LFM Open License v1.0
- Quantization: 8da4w (8-bit dynamic activation / 4-bit weight) + 8-bit embedding (
embedding_quantize: "8,0") - Export: executorch 1.4.0
export_llm, dynamic shape, max_seq_length 2048, XNNPACK extended_ops - Config:
llm_params/lfm2_5_350m_xnnpack_8da4w.yaml
Verification (Mac arm64, 2026-08-13)
Generation gate 3/3 via examples/models/llama/runner/native.py (greedy):
| prompt | output |
|---|---|
| capital of France? | "The capital of France is Paris." |
| 日本の首都は?(日本語) | 「日本の首都は東京です。」 |
| haiku about autumn leaves | 3-line poem, coherent |
Known cosmetic warning: params vocab_size (65536) != tokenizer vocab size.
iPhone 17 Pro (ETBench, XNNPACK CPU, default threads, 2026-08-13)
| metric | value |
|---|---|
| load | 0.5 s |
| ttft (short prompt) | 0.03 s |
| decode | 171-173 tok/s |
Outputs correct (Paris; coherent 128-token story).
Usage note — chat template is required. This is an instruct model: raw untemplated
text makes it emit <|im_end|> immediately (looks like broken generation but is not).
Always wrap prompts as
<|startoftext|><|im_start|>user\n...<|im_end|>\n<|im_start|>assistant\n, eos ids [7].
- Downloads last month
- -