GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
saidutta69
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
saidutta69
Model Tree
Qwen/Qwen3-VL-2B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Full GGUF ladder included - pick the quant that fits your card:
| Your GPU | Recommended quant | Weights |
|---|---|---|
| RTX 4090 / 5090 (24 GB) | Q8_0 | 1.71 GB |
| RTX 4080 / 5080 / 4060 Ti 16G (16 GB) | Q6_K | 1.32 GB |
| RTX 3060 / 4070 / 5070 (12 GB) | Q5_K_M | 1.17 GB |
| RTX 4060 / 3070 (8 GB) | Q4_K_M | 1.03 GB |
| GTX 1660 Super / 2060 / 3050 laptop (6 GB) | IQ4_XS | 0.95 GB |
| CPU-only / Apple Silicon | Q4_K_M | 1.03 GB |
Weights only, at this model's native 2B size; add ~1 GB per 32K of context. Multimodal context carries image tokens too, so budget more KV cache than a text-only model of the same size. OOM? Drop one quant level. Headroom to spare? Go one up.
Trial 74 of a 200-trial Heretic run (seed 1736045380). direction_index was selected per layer.
| Parameter | Value |
|---|---|
| direction_index | per layer |
| attn.o_proj.max_weight | 1.25 |
| attn.o_proj.max_weight_position | 26.80 |
| attn.o_proj.min_weight | 0.97 |
| attn.o_proj.min_weight_distance | 14.91 |
| mlp.down_proj.max_weight | 1.25 |
| mlp.down_proj.max_weight_position | 22.23 |
| mlp.down_proj.min_weight | 0.90 |
| Metric | This model | Original model (Qwen/Qwen3-VL-2B-Instruct) |
|---|---|---|
| KL divergence | 0.0528 | 0 (by definition) |
| Refusals | 5/100 | 97/100 |
KL divergence of 0.0528 with refusals at 5/100 is one of the stronger trade-offs in this batch: the harmful evaluation set drops from 97/100 to 5/100 while the vision tower and grounding behaviour stay close to the original. Small models concentrate refusal behaviour into fewer directions, which is why a 2B edit can land this cleanly.
Fine-tuning a "helpful" persona on top of RLHF'd refusals fights the base model's training and tends to degrade coherence. Abliteration instead finds and edits the specific weight directions responsible for refusal, leaving the rest of the network (and its capabilities) untouched. See the Heretic repo and the original abliteration writeup for the mechanism.
Made with ❤️ by RACER IS OP — follow for more uncensored models
| File | Size |
|---|---|
model.safetensors | 3.96 GB |
BF16, ~2B. The reproduce/ directory carries the full Heretic recipe -
config.toml, requirements.txt, the Optuna study journal, and SHA-256 sums - so this exact
model can be regenerated bit-for-bit. Reproduce it with heretic --reproduce reproduce/reproduce.json.
Full quantization set (14 quants + F16) produced with llama.cpp.
| File | Format | Size |
|---|---|---|
Qwen3-VL-2B-Instruct-heretic-F16.gguf | GGUF F16 | 3.21 GB |
Qwen3-VL-2B-Instruct-heretic-Q2_K.gguf | GGUF Q2_K | 0.72 GB |
Qwen3-VL-2B-Instruct-heretic-IQ3_S.gguf | GGUF IQ3_S | 0.81 GB |
Qwen3-VL-2B-Instruct-heretic-Q3_K_S.gguf | GGUF Q3_K_S | 0.81 GB |
Qwen3-VL architecture (qwen3vl) with its vision encoder - loads natively in llama.cpp / LM Studio / Jan.
Run llama serve -hf saidutta69/Qwen3-VL-2B-Instruct-heretic to pull the default quant.
bash
# llama.cpp - the vision projector ships with the GGUFllama serve -hf saidutta69/Qwen3-VL-2B-Instruct-hereticpython
# transformersfrom transformers import AutoProcessor, AutoModelForImageTextToText model_name = "saidutta69/Qwen3-VL-2B-Instruct-heretic"model = AutoModelForImageTextToText.from_pretrained(model_name, dtype="auto", device_map="auto")processor = AutoProcessor.from_pretrained(model_name) messages = [{"role": "user", "content": [ {"type": "image", "image": "https://example.com/screenshot.png"}, {"type": "text", "text": "Transcribe all text in this image and describe the layout."},]}]inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device)out = model.generate(**inputs, max_new_tokens=1024)print(processor.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))Qwen3-VL emits bounding boxes in a normalised coordinate space, so you can ask for element positions directly and get coordinates back rather than prose descriptions:
python
messages = [{"role": "user", "content": [ {"type": "image", "image": "https://example.com/page.jpg"}, {"type": "text", "text": "Give the bounding box of the submit button."},]}]This is the practical path for UI automation and for building training data from screenshots.
| Architecture | Qwen3VLForConditionalGeneration (vision-language decoder) |
| Parameters | ~2B |
| Layers / heads | 28 layers, 16 attention heads, 8 KV heads, head dim 128 |
| Hidden / intermediate | 2048 / 6144 |
| Position embedding | mRoPE (interleaved, sections 24/20/20), theta = 5,000,000 |
| Context length | 262,144 |
| Vocab | 151,936 |
Qwen3-VL is the vision-language branch of the Qwen series, available in dense and MoE sizes with Instruct and Thinking editions. This generation brings deeper visual perception and reasoning, extended context, stronger spatial and video understanding, and improved agent interaction — OCR, grounding, and video temporal reasoning are the headline capabilities over the earlier Qwen2.5-VL.
Refusal suppression is deliberate and works as intended: this model will comply with requests the base model would refuse, including some it shouldn't. There is no safety filtering layered on top. You are responsible for how you deploy it — don't put this behind an unmoderated public-facing endpoint serving third parties. It inherits Qwen3-VL's factual limitations and biases; abliteration removes refusal directions, it doesn't add capability or judgment.
Inherits the apache-2.0 license from the base model.
| mlp.down_proj.min_weight_distance | 6.95 |
Qwen3-VL-2B-Instruct-heretic-Q3_K_M.gguf | GGUF Q3_K_M | 0.88 GB |
Qwen3-VL-2B-Instruct-heretic-Q3_K_L.gguf | GGUF Q3_K_L | 0.93 GB |
Qwen3-VL-2B-Instruct-heretic-IQ4_XS.gguf | GGUF IQ4_XS | 0.95 GB |
Qwen3-VL-2B-Instruct-heretic-Q4_K_S.gguf | GGUF Q4_K_S | 0.99 GB |
Qwen3-VL-2B-Instruct-heretic-Q4_0.gguf | GGUF Q4_0 | 0.98 GB |
Qwen3-VL-2B-Instruct-heretic-Q4_1.gguf | GGUF Q4_1 | 1.06 GB |
Qwen3-VL-2B-Instruct-heretic-Q4_K_M.gguf | GGUF Q4_K_M | 1.03 GB |
Qwen3-VL-2B-Instruct-heretic-Q5_K_S.gguf | GGUF Q5_K_S | 1.15 GB |
Qwen3-VL-2B-Instruct-heretic-Q5_K_M.gguf | GGUF Q5_K_M | 1.17 GB |
Qwen3-VL-2B-Instruct-heretic-Q6_K.gguf | GGUF Q6_K | 1.32 GB |
Qwen3-VL-2B-Instruct-heretic-Q8_0.gguf | GGUF Q8_0 | 1.71 GB |
| bfloat16 |
| Modalities | Text and image/video in, text out |
| Base model | Qwen/Qwen3-VL-2B-Instruct |