Downloads · 30 days
1.6K
33% of all-time downloads
mmp2055/Qwen2.5-VL-3B-uncensored-bughunter-v9
Qwen2.5-VL-3B-uncensored-bughunter-v9 is a image-text-to-text model from mmp2055. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for gguf. The card lists the license as apache-2.0.
On-device agent for mobile bug-bounty workflows. Light LoRA fine-tune of huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated to steer the researcher persona while preserving the base's native tool-calling and non-refusal beh…
Downloads · 30 days
1.6K
33% of all-time downloads
All-time downloads
4.8K
Public
Repo size
8.7 GB
Likes
7
Public
Click a slice to open those files.
.gguf3.3 GB · 98%
From the Hugging Face model README
On-device agent for mobile bug-bounty workflows. Light LoRA fine-tune of huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated to steer the researcher persona while preserving the base's native tool-calling and non-refusal behavior.
Positioned as a tool orchestrator, NOT a standalone knowledge base. Total Q4_K_M footprint: ~1.9 GB, ~2.5 GB active RAM, runs at 5-40 tok/s on modern phones via llama.cpp bindings.
This is a 3B mobile model with a light fine-tune (1 epoch, LoRA rank 8, lr 2e-5). Empirically validated behavior:
| Result | |
|---|---|
Native tool-calling (<tool_call>{name, arguments}</tool_call>) | ✅ Works — emits well-formed JSON, picks appropriate tools, sets finish_reason: tool_calls |
| Uncensored (no categorical refusals) | ✅ Inherited from huihui-ai base |
| Vision (image input) | ✅ Preserved from base (mmproj-BF16.gguf, frozen during fine-tune) |
| Follows system prompt (researcher persona) | ✅ Respects role |
Chat template with {% if tools %} block | ✅ Preserved verbatim from base (2,427 chars) |
| Deep offensive-security payload knowledge | ⚠️ Limited — 3B parameters + light LoRA cannot hold the depth needed to reliably emit correct XSS/SQLi/etc payloads. Expect generic or occasionally incorrect content on direct payload queries. |
| Standalone offline oracle | ❌ Use v8 (35B MoE) for depth queries |
kid + algorithm confusion, gadget-chain construction).v9The HF slot mmp2055/Qwen2.5-VL-3B-uncensored-bughunter-v9 was re-trained. Concrete changes:
Qwen/Qwen2.5-VL-3B-Instruct (needed a full abliteration pass) to huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated (pre-abliterated + tool-calling chat template with the {% if tools %} block).v9 patched an "aggressive researcher" template that broke tool-calling and induced generation loops.additions_v9.2 use "You are a professional bug bounty researcher with access to tools" (was leaking "You are a helpful assistant with access to tools" in the previous iteration, contaminating the researcher persona).Qwen2.5-VL-3B-uncensored-bughunter-v9.Q4_K_M.gguf — the LLM (~1.9 GB, Q4_K_M).mmproj-BF16.gguf — vision projector (~1.3 GB, BF16). Only needed for image input.Modelfile — Ollama-compatible template with baked-in temperature=0.1 and researcher SYSTEM prompt./adapter — LoRA weights (rank 8, alpha 16) for anyone building on top.| Parameter | Value | Why |
|---|---|---|
temperature | 0.1 | Very deterministic. Higher temps reintroduce base residual behavior (defensive answers, less consistent tool-call emission). |
top_p | 0.9 | Standard nucleus sampling. |
top_k | 40 | Standard. |
repeat_penalty | 1.1 | Prevents loops on longer answers. |
num_ctx (context) | 4,096-8,192 on mobile; up to 32,768 on desktop | Memory budget on-device. |
The bundled Modelfile bakes in temperature=0.1. In LM Studio / llama-server set it manually — the runtime defaults (0.7-0.8) are too high for this fine-tune.
The GGUF preserves the huihui-ai chat template including the {% if tools %} block. Supply an OpenAI-style tools array in the request:
POST /v1/chat/completions
{
"model": "qwen2.5-vl-3b-uncensored-bughunter-v9",
"messages": [
{"role": "system", "content": "You are a bug bounty researcher with tool access."},
{"role": "user", "content": "Fetch the security.txt of https://example.com and summarize it."}
],
"tools": [
{
"type": "function",
"function": {
"name": "http_fetch",
"description": "Fetch a URL and return response body",
"parameters": {
"type": "object",
"properties": {"url": {"type": "string"}},
"required": ["url"]
}
}
}
],
"temperature": 0.1,
"max_tokens": 512
}
Response (verified on this model):
{
"choices": [{
"finish_reason": "tool_calls",
"message": {
"role": "assistant",
"content": "",
"tool_calls": [{
"type": "function",
"function": {
"name": "http_fetch",
"arguments": "{\"url\":\"https://example.com/security.txt\"}"
}
}]
}
}]
}
The model returns clean JSON in tool_calls. Wire the response to your executor and pass the result back with role: "tool" in the next turn.
.gguf files into the same folder: ~/.lmstudio/models/mmp2055/Qwen2.5-VL-3B-uncensored-bughunter-v9-GGUF/.llama-server \
-m Qwen2.5-VL-3B-uncensored-bughunter-v9.Q4_K_M.gguf \
--mmproj mmproj-BF16.gguf \
-c 8192 \
-fa on \
-ngl 999 \
--port 8080
OpenAI-compatible endpoint at http://127.0.0.1:8080/v1/chat/completions.
ollama create bughunter-v9 -f Modelfile
ollama run bughunter-v9
The Modelfile supplies a plain ChatML TEMPLATE and the researcher SYSTEM prompt. Ollama's Go-based Jinja parser cannot fully evaluate the Qwen2.5-VL template constructs (namespace(), <|vision_start|>, {% for content in message['content'] %}) so the plain-ChatML fallback is needed. LM Studio and llama.cpp both use the embedded template correctly without the Modelfile.
Base: huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated. Pre-abliterated variant of Qwen2.5-VL-3B-Instruct with tool-calling chat template preserved.
Data: 15,914 ShareGPT examples = 14,914 from the v6/v7/v8 private corpus + 1,000 additions (500 tool_call examples with the researcher system prompt, 300 general researcher queries, 200 anti-collapse). Corpus is private.
Hyperparameters — deliberately light to avoid overwriting base behavior:
| Parameter | Value |
|---|---|
| LoRA rank | 8 |
| LoRA alpha | 16 |
| LoRA dropout | 0.05 |
| Target modules | q/k/v/o_proj + gate/up/down_proj (text tower only) |
| Vision layers | FROZEN (finetune_vision_layers=False) |
| Learning rate | 2e-5 |
| Epochs | 1 |
| Effective batch | 8 (per_device 4 × grad_accum 2 on A100) |
| Max seq length | 2,048 (auto-reduced by Unsloth from 4,096) |
| Scheduler | Cosine with 3% warmup |
| Precision | bf16 |
| Training quantization | QLoRA 4-bit (bitsandbytes) |
| Total training steps | 2,429 |
| Trainable parameters | 14,966,784 (0.40% of 3.77B total) |
| Runtime | ~78 min on A100 40 GB |
Framework: Unsloth FastVisionModel 2026.9.4 + transformers 5.5.0 + peft 0.20.0 + trl 0.24.0 on Colab Pro A100.
Training metrics (eval every ~250 steps):
| Step | train_loss | eval_loss |
|---|---|---|
| 250 | 1.303 | 1.548 |
| 500 | 1.273 | 1.505 |
| 750 | 1.205 | 1.473 |
| 1000 | 1.257 | 1.447 |
| 1250 | 1.161 | 1.428 |
| 1500 | 1.197 | 1.415 |
| 1750 | 1.166 | 1.409 |
| 2000 | 1.237 | 1.405 |
| 2250 | 1.169 | 1.403 |
| 2429 (final) | 1.228 | 1.403 |
Clean descent, no overfit, plateau reached around step 2,000 — expected for 1 epoch on 15k examples with a small adapter.
LoRA strategy (text tower only):
q/k/v/o_proj (attention) + gate/up/down_proj (MLP) of every one of the 36 decoder layers of the language model.visual.* and merger.* — vision encoder + projector kept intact so the base's image understanding is preserved.Chat template: preserved verbatim from huihui-ai base (2,427 chars, includes the {% if tools %} block for Qwen2.5 native tool-calling). Not patched, not edited.
convert_hf_to_gguf.py + llama-quantize (llama.cpp master, Sep 2026).huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated since vision was frozen).transformers 5.5.0, the tokenizer_config.json saved by Unsloth's merge triggers AttributeError: 'list' object has no attribute 'keys' in convert_hf_to_gguf.py. Workaround (applied): copy tokenizer files from the base model before running the converter./adapterThe LoRA adapter used to train this model is in this repo under /adapter:
adapter/
├── adapter_config.json # PEFT config: rank 8, alpha 16, dropout 0.05
├── adapter_model.safetensors # LoRA delta weights (~30 MB)
├── chat_template.jinja # 2,427 chars, preserved from base
├── tokenizer_config.json
├── tokenizer.json
└── README.md
Merge into the base:
import torch
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from peft import PeftModel
from huggingface_hub import snapshot_download
BASE = "huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated"
adapter_dir = snapshot_download(
"mmp2055/Qwen2.5-VL-3B-uncensored-bughunter-v9",
allow_patterns="adapter/*",
)
base = Qwen2_5_VLForConditionalGeneration.from_pretrained(
BASE, torch_dtype=torch.bfloat16, device_map="cpu", trust_remote_code=True
)
model = PeftModel.from_pretrained(base, f"{adapter_dir}/adapter")
model = model.merge_and_unload()
processor = AutoProcessor.from_pretrained(BASE, trust_remote_code=True)
kid traversal chains, POP gadget chain construction, cross-chain signature replay — same limitation as previous iterations (dataset gap inherited).Distributed for:
Not distributed for:
The user assumes all legal responsibility for use of the model.
FastVisionModel.@misc{qwen25vl_bughunter_v9_2,
title = {Qwen2.5-VL-3B-uncensored-bughunter-v9 (v9.2 iteration): mobile tool-caller for on-device bug bounty workflows},
year = {2026},
base = {huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated},
method = {Light LoRA fine-tune (rank=8, alpha=16, lr=2e-5, 1 epoch, QLoRA 4-bit, Unsloth FastVisionModel) on text tower only, vision frozen. No secondary abliteration (inherited from base). Chat_template preserved verbatim from base including {% if tools %} block for Qwen2.5 native tool-calling.},
data = {15,914 ShareGPT examples: 14,914 from prior v6-v8 private corpus + 1,000 additions (500 tool_call, 300 general researcher queries, 200 anti-collapse) with corrected system prompts.}
}