Downloads · 30 days
27
5% of all-time downloads
LFM2.5-1.2B-Thinking-Autoround-Safetensors
simaai/LFM2.5-1.2B-Thinking-Autoround-Safetensors
LFM2.5-1.2B-Thinking-Autoround-Safetensors is a text generation model from simaai. Use it when you need the model to write or continue text. It is set up for transformers.
This local, pre-LLiMa Hugging Face safetensors checkpoint quantizes LiquidAI/LFM2.5-1.2B-Thinking for the LLiMa/Sima.ai pipeline. The immutable source revision was not captured in the local cache and must be pinned be…
All-time downloads
553
Public
Parameters
1.3B
863 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors863 MB · 99%
Parameter types
How the weights are stored.
I321.2B · 90%
Try a prompt
Base models
Model card
From the Hugging Face model README
LFM2.5-1.2B-Thinking Autoround Safetensors
This local, pre-LLiMa Hugging Face safetensors checkpoint quantizes LiquidAI/LFM2.5-1.2B-Thinking for the LLiMa/Sima.ai pipeline. The immutable source revision was not captured in the local cache and must be pinned before any publication.
Quantization
| Component | Method | Format | Details |
|---|---|---|---|
| Decoder and short-convolution projection Linear layers | AutoRound | symmetric INT4, G256 | 92 modules; lm_head excluded; W4A16 |
lm_head | GPTQ | symmetric INT4, G256 | static actorder; block size 128; dampening 0.01 |
| Short convolution kernels | — | BF16 | Conv1d weights were not quantized |
| Activations | — | FP16/BF16 | not activation-quantized |
Calibration used 512 fixed-order samples from NeelNanda/pile-10k, sequence length 1024, AutoRound 200 iterations, batch size 1, and device ID 0.
Evaluation
Full wikitext-2-raw-v1 was evaluated through lm-eval's wikitext task with no example limit and matched settings.
| Checkpoint | Word perplexity | Status |
|---|---|---|
LiquidAI/LFM2.5-1.2B-Thinking | 48.905688365273 | Full run: 2026-07-15 |
| This W4 decoder + GPTQ W4/G256 head checkpoint | 55.149647000135 | +6.243958634862 / +12.767346383568% |
Raw JSON evidence is retained in perplexity_results/full_wikitext/ in this workspace.
Reproduction
Included files: README.md, quantize.py, recipe.yaml, and versions.txt.
python quantize.py --model-path /path/to/LFM2.5-1.2B-Thinking --output-dir /path/to/new-output-directory
Quantization completed on an NVIDIA A100 80 GB GPU on 2026-07-15; saved scales passed the NaN/Inf check. The compressed-checkpoint smoke test passed: 93 compressed modules including lm_head loaded on CUDA and completed greedy generation.
Deployment and limitations
This checkpoint is intended for subsequent LLiMa compilation and Sima.ai evaluation. It has not been uploaded automatically. Quantization quality can vary by language, domain, prompt format, and context length; validate the compiled deployment for the intended workload.
- Task
- Text Generation
- Library
- transformers
- Type
- lfm2
- Created
- Jul 16, 2026
- Updated
- Sep 1, 2026