Skip to content
Other

coder-lfm3-1.7b-qwen4-max

compact coding model based on LFM3

2 min readupdated

This is a fine-tuned version of LiquidAI/LFM3-1.7B (1.73B parameters) built specifically for programming, computer science, and statistics. Creator AJ-Gonzalez trained it on reasoning traces and code samples from the flagship qwen/qwen4-max model to instill step-by-step technical problem-solving skills into a compact architecture. Compared to the base LFM3-1.7B, this release delivers noticeably higher accuracy on code generation benchmarks and handles complex repository contexts better. It is a solid fit for developers looking to run local autocomplete and script generation on low-end hardware or Apple Silicon laptops.

Key features

  • Built on the LFM3-1.7B architecture and distilled with reasoning traces and code blocks from Qwen4 Max.
  • Shows solid gains on the HumanEval+ benchmark: base pass@1 jumped from 0.695 to 0.732 (+3.7 points), while the harder split rose from 0.652 to 0.683.
  • Ships with ready-to-use weights for different setups: MLX 4-bit (946 MB), GGUF Q4_K_M (1.1 GB), the full bf16 checkpoint (3.3 GB), and a standalone PEFT LoRA adapter (63 MB).
  • On a base M1 MacBook with 8 GB of RAM, the MLX 4-bit build pushes around 55 tokens per second while using only ~1.15 GB of memory.

What you can do

  • Spin up a local coding assistant on a Mac via mlx_lm to write Python functions, implement algorithms, and solve statistics problems.
  • Load the GGUF build into llama.cpp or compatible local frontends for offline code suggestions on a standard PC without a beefy GPU.
  • Drop the standalone LoRA adapter (63 MB) into existing LFM3-1.7B pipelines using Transformers.

Example

mlx_lm.generate --model AJ-Gonzalez/coder-lfm3-1.7b-qwen4-max \
    --prompt "Write a Python function that returns the first 12 prime numbers." \
    --max-tokens 3000
Read moreDetails
Training: 4-bit NF4 QLoRA (rank 16, alpha 32, 16.47M trainable parameters) trained on 70 examples (340 training chunks of 2048 tokens) in 50.2 minutes on a single T4 GPU.
SWE-bench prompt NLL metrics: perplexity (ppl) dropped from 12.34 to 9.31 on T4 fp16; in MLX 4-bit, NLL improved from 2.6936 to 2.3156.
Limitations: 4-bit quantization may slightly alter output quality compared to fp16; generated code should always be reviewed.
License: LFM Open License v1.0 (free for commercial use up to $10M in annual revenue).
Discuss Telegram 15 13.4K

Will it run?

Runs in U32

Fits without quantisation.

  • U32weights 3.2 GB · ready files4.0 GB
  • FP8 / INT8recommendedweights 1.6 GB2.3 GB
  • GGUF Q4_K_Mweights 1.0 GB · ready files1.6 GB
  • 4-bitweights 0.9 GB1.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · U32 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # AJ-Gonzalez/coder-lfm3-1.7b-qwen4-max
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download AJ-Gonzalez/coder-lfm3-1.7b-qwen4-max --local-dir ~/models/coder-lfm3-1.7b-qwen4-max
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. LFM3-1.7B-Basefamily root
  2. LFM3-1.7B
  3. coder-lfm3-1.7b-qwen4-max

New posts about versions of LFM3-1.7B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 9All on Hugging Face

About the model

AJ-Gonzalez/coder-lfm3-1.7b-qwen4-max
Author
AJ-Gonzalez
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

13.4K15

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Language models

xevyo-01

xevyo-01