dawncr0w/crowqwen3.6-35b-a3b-agent-heretic-gguf overview
crowqwen3.6 35b a3b agent heretic gguf CrowQwen is a Qwen3.6 35B A3B derivative tuned for local coding agents: repository edits, structured tool calls, and con…
Runs locally from ~15.61 GB disk (16 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| crowqwen3.6-35b-a3b-agent-heretic-Q3_K_M.gguf | GGUF | Q3_K_M | 15.61 GB | Download |
| crowqwen3.6-35b-a3b-agent-heretic-Q4_K_M.gguf | GGUF | Q4_K_M | 19.71 GB | Download |
| crowqwen3.6-35b-a3b-agent-heretic-Q5_K_M.gguf | GGUF | Q5_K_M | 23.03 GB | Download |
| crowqwen3.6-35b-a3b-agent-heretic-Q6_K.gguf | GGUF | Q6_K | 26.56 GB | Download |
| crowqwen3.6-35b-a3b-agent-heretic-Q8_0.gguf | GGUF | Q8_0 | 34.37 GB | Download |
Model Details
| Model ID | dawncr0w/crowqwen3.6-35b-a3b-agent-heretic-gguf |
|---|---|
| Author | dawncr0w |
| Pipeline | text-generation |
| License | apache-2.0 |
| Base model | Qwen/Qwen3.6-35B-A3B |
| Last modified | 2026-06-26T06:29:34.000Z |
Model README
---
license: apache-2.0
base_model:
- Qwen/Qwen3.6-35B-A3B
library_name: llama.cpp
pipeline_tag: text-generation
tags:
- gguf
- llama.cpp
- qwen
- qwen3.6
- moe
- coding
- agent
- tool-use
- heretic
- refusal-reduced
---
crowqwen3.6-35b-a3b-agent-heretic-gguf
CrowQwen is a Qwen3.6-35B-A3B derivative tuned for local coding agents:
repository edits, structured tool calls, and concise execution loops. This repo
contains GGUF builds for llama.cpp-compatible text inference.
Why Pick This
- Agentic-coding and tool-use tuning on top of Qwen3.6-35B-A3B.
- Local validation improved tool-call reliability and average response time.
- Heretic refusal-direction reduction applied after SFT.
Files
| File | Use when |
| --- | --- |
| Q8_0 | Highest-quality GGUF in this release |
| Q6_K | Strong quality with smaller size |
| Q5_K_M | Balanced local deployment |
| Q4_K_M | Lower memory use |
| Q3_K_M | Smallest GGUF in this release |
Release Check
Compared against Qwen3.6-35B-A3B with matching local settings before GGUF
packaging.
| Check | Same-quant base | CrowQwen | Delta |
| --- | ---: | ---: | ---: |
| Agentic coding full pass | 38/56 | 41/56 | +3 |
| Tool-use pass | 44/48 | 48/48 | +4 |
| Tool-use valid JSON | 46/48 | 48/48 | +2 |
| Agentic avg time | 7.135s | 6.101s | Faster |
| Tool-use avg time | 1.109s | 0.772s | Faster |
These are local release checks, not broad leaderboard benchmarks.
Use
llama-server \
-m crowqwen3.6-35b-a3b-agent-heretic-Q8_0.gguf \
--host 127.0.0.1 \
--port 8000
from openai import OpenAI
client = OpenAI(base_url="http://127.0.0.1:8000/v1", api_key="local")
response = client.chat.completions.create(
model="crowqwen3.6-35b-a3b-agent-heretic-gguf",
messages=[{"role": "user", "content": "Review this patch for bugs."}],
)
print(response.choices[0].message.content)
GGUF builds are exported as the text trunk (--no-mtp) for current llama.cpp
compatibility. Use the MLX/oMLX repo for MTP/vision-preserving artifacts.
Based on Qwen/Qwen3.6-35B-A3B.
Run dawncr0w/crowqwen3.6-35b-a3b-agent-heretic-gguf with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models