lfm3-1.7b-sweep-sft (work in progress)

Experimental fine-tunes of LiquidAI/LFM3-1.7B-Base for code completion and Zed-style next-edit prediction. This is a modified version of LFM3-1.7B-Base: all model weights in this repository were changed by fine-tuning. It is distributed under the LFM Open License v1.0 (see LICENSE), including its commercial-use threshold. Not affiliated with or endorsed by Liquid AI.

Each training run is a git tag; load one with AutoModelForCausalLM.from_pretrained("Walltim3/lfm3-1.7b-sweep-sft", revision="<tag>", trust_remote_code=True). checkpoints/ holds in-progress training state.

Tag What Training data License notes
code-sft chat-format code SFT (experiment, superseded) 36k KodCode-V1 solutions KodCode is CC BY-NC 4.0: this variant is for non-commercial use only
code-fim (planned) raw code + fill-in-the-middle (Zed qwen layout) permissively licensed files from bigcode/commitpack LFM Open License v1.0
sweep-sft-* (planned) next-edit in Zed's sweep prompt format permissively licensed commits from bigcode/commitpack LFM Open License v1.0

Experimental; no guarantees of quality or fitness for any purpose.

Downloads last month
2,937
Safetensors
Model size
2B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Walltim3/lfm3-1.7b-sweep-sft

Quantized
(21)
this model