Skip to content
Other

LifeMate V7 Weights

LoRA adapters and soup builds based on Qwen3.5-4B

2 min read

Creator Nhat42 has bundled a set of LoRA adapters (lightweight fine-tuning add-ons), merged LoRA soup builds, and evaluation logs for the LifeMate V7 system into a single repository. It is built on the Qwen/Qwen3.5-4B language model, trained on synthetic data to generate responses and tool calls. The project's standout feature is its exact rank concatenation method (exact_rank_concat), which merges up to five different adapters into one without precision loss—even on a standard CPU. This release is useful for developers experimenting with model weight merging and building compact LLM agents under 5B parameters.

Key features

  • Pre-merged g_all5 build: combines five specialized adapters (F1–F5) with equal weights, boosting the final score on unseen validation to 73.89 compared to 67.54 for the baseline.
  • CPU-only merging without a GPU: the repository documents a reproducible workflow for merging individual LoRA modules into a single file with SHA256 checksum verification.
  • Task-specialized adapters: separate modules in the adapters/ folder were fine-tuned for distinct subtasks—compute, date handling, logic branching, and DPO alignment.
  • Transparent benchmarks: detailed evaluation reports (summary.json, details.jsonl) are included for 17 configurations across the val_seen and val_unseen splits.

What you can do

  • Build a compact Qwen3.5-4B-based model for local agents that need to call external functions and APIs.
  • Use the ready-made adapters from the adapters/ directory to experiment with model merging and compare different fine-tuning strategies.

Example

pip install huggingface_hub
export HF_TOKEN=hf_xxx
huggingface-cli download Nhat42/lifemate-v7-weights \
    --include "submission/g_all5/*" \
    --local-dir outputs/v7/hf_weights
Read moreDetails
Base model: Qwen/Qwen3.5-4B (final merged build under 5B parameters)
Best result (g_all5): final_score 73.89 (hash_match 82.33%) on val_unseen (232 samples) and 85.46 (92.22%) on val_seen (257 samples)
LoRA parameters: r=32, lora_alpha=64 for single adapters (232.6 MB each) and r=160, lora_alpha=320 for g_all5 (1162.6 MB)
Note: the repository is marked as Private on the model card; an access token is provided for CLI downloads
Discuss Telegram 0

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # Nhat42/lifemate-v7-weights
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 Nhat42/lifemate-v7-weights --local-dir ~/models/lifemate-v7-weights
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to lifemate-v7-weights
  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

New posts about versions of lifemate-v7-weights (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Nhat42/lifemate-v7-weights
Author
Nhat42
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Other

Cloudie

Cloudie