AlexWortega/qwen35-4b-clawd-rift-gguf

Qwen3.5-4B Clawd-RIFT — GGUF quantizations

Usage with llama.cpp

llama-cli -m clawd-rift-Q5_K_M.gguf -p 'Your prompt'
llama-server -m clawd-rift-Q5_K_M.gguf --port 8080

In Ollama / LM Studio: import GGUF directly. Set chat template to Hermes-style with <tool_call>{json}</tool_call> for tool use.

Evaluation results

tbench-2 (89 docker tasks via Pi-style runner)

7/89 (7.9%). Tasks unique to clawd-rift: fix-ocaml-gc, pytorch-model-recovery.

ClawGym-Bench (200 tasks via openclaw scaffold)

Comparison to RUC-AIBOX ClawGym leaderboard (compact open-weight models):

Optimal inference parameters

Sampling sweet-spot is scaffold-dependent.

Universal default that loses only ~5% on each:

temperature=0.5, top_p=0.95, top_k=40, repetition_penalty=1.05

Training methodology — pipeline of 3 stages

Stage 1: Soyuz SFT (ckpt600 — base agent format)

QLoRA r=64 alpha=128 on Qwen/Qwen3.5-4B.

• —Datasets: `AlexWortega/Soyuz-sft` + `AlexWortega/AgentTrove`
• —Format: Hermes-style JSON tool calls (<tool_call>{"name":...,"arguments":...}</tool_call>)
• —600 steps total, seq=8K, Muon optimizer for LoRA matrices
• —Output: ckpt-400, ckpt-600 (intermediate); soup_sum = ckpt400 + ckpt600 (arithmetic merge)

Stage 2: ClawGym continue-train (clawd-100, clawd-200 — openclaw scaffold adaptation)

Continue-train ckpt600 on filtered `RUC-AIBOX/ClawGym-Trajectory`.

• —1937 trajectories (filtered ≤16K tokens out of 24.5K)
• —200 steps, seq=16K, LR=1e-4, AdamW
• —Hermes chat template + openclaw native tools (read/write/exec/web_search/...)
• —Output: clawd-100 (mid), clawd-200 (final)

Stage 3: RIFT — own rollouts + reward feedback

True RIFT loss on top of clawd-200:

# positive (reward > 0): NLL × reward — weighted SFT
# negative (reward = 0): exp(logp) × negative_scale — unlikelihood

• —61 trajectories from soup_sum's own ClawGym rollouts (46 pos + 15 neg, reward 0-1)
• —5 epochs / 80 steps, LR=2e-5
• —Implementation: `compare_offlinegpro/src/trainers/offline_losses.py`
• —Output: clawd-rift ← this model

Repos

GGUF breakdown:

• —clawd-rift-f16.gguf (7.9 GB, baseline)
• —clawd-rift-Q8_0.gguf (4.2 GB, near-lossless)
• —clawd-rift-Q5_K_M.gguf (2.9 GB, recommended)
• —clawd-rift-Q4_K_M.gguf (2.6 GB, smallest)

W&B training logs: https://wandb.ai/alexwortega/vae-llm-agents

Related: Stage-1-only model (qwen35-4b-soyuz)

A cleaner, stronger reference for the Stage-1 base (Soyuz SFT only — no ClawGym, no RIFT) is now available, trained as full bf16 LoRA r=128 (vs QLoRA r=64 here):

Final eval on Soyuz-clean held-out: loss=0.247, token_acc=0.936. Trained on the cleaned 11-stream subset of `AlexWortega/Soyuz-sft` at seq=16K, 1 epoch.

Useful if you want only the Hermes-tool-call SFT without the ClawGym/RIFT specialization.

Loading caveat

These GGUF files were converted by an older llama.cpp build before upstream support for the Qwen3.5 hybrid linear+full attention architecture stabilized. Some llama.cpp builds may complain about missing tensor or unsupported architecture when loading. The merged HF weights at qwen35-4b-clawd-rift-merged are the canonical reference.