QwenPaw-Flash-9B-heretic MTP Version: [QwenPaw-Flash-9B-heretic-MTP-GGUF ]

🏆 BenchLocal Total: 3873/5000 (77.5%) — Outperforms 35B MoE models

Uncensored · Abliterated · Agent-Optimized

Uncensored version of QwenPaw-Flash-9B, processed with Heretic v1.3.0 abliteration. Fine-tuned from Qwen3.5-9B for autonomous agent scenarios.

Test Environment: NVIDIA RTX 5070 Ti (16GB) · llama.cpp (turboquant build) · Q6_K quant

Framework: BenchLocal — local model agent evaluation suite

Methodology: Each scenario run once, no retries, no second attempts

🛠️ ToolCall-15 — Tool Calling Stability (93.3%)

🤖 HermesAgent-20 — Complex Agent Tasks (77.2%)

4-Model Comparison (9B vs 35B / 26B):

🐛 BugFind-15 — Code Debugging (61.9%)

• Base model**: QwenPaw-Flash-9B (Qwen3.5-9B fine-tuned for QwenPaw autonomous agent scenarios)
• Tool**: Heretic v1.3.0 (automatic directional ablation)
• Best trial**: #194 / 230 trials

direction_index = 21.13 attn.o_proj.max_weight = 1.42 attn.o_proj.max_weight_position = 21.72 attn.o_proj.min_weight = 1.11 attn.o_proj.min_weight_distance = 18.14 mlp.down_proj.max_weight = 1.48 mlp.down_proj.max_weight_position = 21.23 mlp.down_proj.min_weight = 1.47 mlp.down_proj.min_weight_distance = 17.47

• Quantization**: BNB_4BIT (during training)
• Batch size**: 32
• Trials**: 230 (final selection: Trial #194)
• Datasets**: mlabonne/harmless_alpaca + mlabonne/harmful_behaviors

• Type**: Qwen3_5ForConditionalGeneration (multimodal with vision encoder)
• Parameters**: ~9B
• Layers**: 32 (hybrid: Gated DeltaNet + Gated Attention)
• Context length**: 262,144 tokens

llama.cpp

llama-server -m QwenPaw-Flash-9B-heretic-Q6_K.gguf -ngl 99 -fa on -c 8192 --host 0.0.0.0 --port 8088

LM Studio

Load the GGUF file directly.

1. Safety filters have been significantly reduced via abliteration
2. KL divergence is only 0.0225 — minimal impact on model intelligence
3. The original model supports multimodal (vision); GGUF versions require the mmproj file
4. BenchLocal scores measured at Q6_K on RTX 5070 Ti 16GB with llama.cpp (turboquant). Each scenario was run once with no retries — scores represent single-shot performance
5. Outperforms Qwen3.6-35B-A3B (MoE, Thinking ON) and Gemma-4-26B in HermesAgent-20 despite being 1/4 the parameter count
6. TC-11 (Prompt boundary) and TC-12 (Ambiguous request) fixed — areas where the 35B APEX model failed
7. Primary weakness: BugFind-15 false positives — the abliterated model tends to over-eagerly "fix" correct code
8. Please use responsibly

• Heretic — Automated censorship removal
• agentscope-ai/QwenPaw-Flash-9B — Base model
• llama.cpp — GGUF quantization and inference
• BenchLocal — Local model agent evaluation suite