Qwen3.5-9B-The-Bradbury-F451-Pro-Writer-Uncensored-Heretic

• Vision (images) tested -> working with new training.

Benchmarks:

Qwen3.5-9B-The-Bradbury-F451-Pro-Writer-Uncensored-Heretic
           arc   arc/e boolq hswag obkqa piqa  wino
mxfp8      0.590,0.761,0.883,0.697,0.444,0.779,0.684

Qwen3.5-9B
           arc   arc/e boolq hswag obkqa piqa  wino
mxfp8      0.571,0.719,0.895,0.683,0.426,0.770,0.671

Performance

• Suggest min q4ks (non-imatrix) or IQ3S (imatrix).
• Tested with rep pen of 1 (off).
• Context: 256k (default).

• Other versions in testing.
• Information from Qwen's repo below.
• Video portions of the model were NOT TESTED.

• Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
• If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.

Qwen3.5-9B

[!Note] This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.

These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc.

Qwen3.5 Highlights

• Unified Vision-Language Foundation: Early fusion training on multimodal tokens achieves cross-generational parity with Qwen3 and outperforms Qwen3-VL models across reasoning, coding, agents, and visual understanding benchmarks.
• Efficient Hybrid Architecture: Gated Delta Networks combined with sparse Mixture-of-Experts deliver high-throughput inference with minimal latency and cost overhead.
• Scalable RL Generalization: Reinforcement learning scaled across million-agent environments with progressively complex task distributions for robust real-world adaptability.
• Global Linguistic Coverage: Expanded support to 201 languages and dialects, enabling inclusive, worldwide deployment with nuanced cultural and regional understanding.
• Next-Generation Training Infrastructure: Near-100% multimodal training efficiency compared to text-only training and asynchronous RL frameworks supporting massive-scale agent scaffolds and environment orchestration.

Model Overview

• Type: Causal Language Model with Vision Encoder
• Training Stage: Pre-training & Post-training
• Language Model
  • Number of Parameters: 9B
  • Hidden Dimension: 4096
  • Token Embedding: 248320 (Padded)
  • Number of Layers: 32
  • Hidden Layout: 8 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
  • Gated DeltaNet:
    • Number of Linear Attention Heads: 32 for V and 16 for QK
    • Head Dimension: 128
  • Gated Attention:
    • Number of Attention Heads: 16 for Q and 4 for KV
    • Head Dimension: 256
    • Rotary Position Embedding Dimension: 64
  • Feed Forward Network:
    • Intermediate Dimension: 12288
  • LM Output: 248320 (Padded)
  • MTP: trained with multi-steps
• Context Length: 262,144 natively and extensible up to 1,010,000 tokens.

Benchmark Results

Language

Vision Language

Quickstart

[!Important] Qwen3.5 models operate in thinking mode by default, generating thinking content signified by <think>\n...</think>\n\n before producing the final responses. To disable thinking content and obtain direct response, refer to the examples here.

Serving Qwen3.5

[!Important] Inference efficiency and throughput vary significantly across frameworks. We recommend using the latest framework versions to ensure optimal performance and compatibility. For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.

[!Important] The model has a default context length of 262,144 tokens. If you encounter out-of-memory (OOM) errors, consider reducing the context window. However, because Qwen3.5 leverages extended context for complex tasks, we advise maintaining a context length of at least 128K tokens to preserve thinking capabilities.

SGLang

• Standard Version: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
  python -m sglang.launch_server --model-path Qwen/Qwen3.5-9B --port 8000 --tp-size 1 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3
• Tool Use: To support tool use, you can use the following command.
  python -m sglang.launch_server --model-path Qwen/Qwen3.5-9B --port 8000 --tp-size 1 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --tool-call-parser qwen3_coder
• Multi-Token Prediction (MTP): The following command is recommended for MTP:
  python -m sglang.launch_server --model-path Qwen/Qwen3.5-9B --port 8000 --tp-size 1 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --speculative-algo NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4

vLLM

• Standard Version: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
  vllm serve Qwen/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --max-model-len 262144 --reasoning-parser qwen3
• Tool Call: To support tool use, you can use the following command.
  vllm serve Qwen/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder
• Multi-Token Prediction (MTP): The following command is recommended for MTP:
  vllm serve Qwen/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --max-model-len 262144 --reasoning-parser qwen3 --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}'
• Text-Only: The following command skips the vision encoder and multimodal profiling to free up memory for additional KV cache:
  vllm serve Qwen/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only

KTransformers

Hugging Face Transformers

Using Qwen3.5 via the Chat Completions API

[!Tip] We recommend using the following set of sampling parameters for generation

• Thinking mode for general tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
• Thinking mode for precise coding tasks (e.g. WebDev): temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
• Instruct (or non-thinking) mode for general tasks: temperature=0.7, top_p=0.8, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
• Instruct (or non-thinking) mode for reasoning tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0

Please note that the support for sampling parameters varies according to inference frameworks.

Text-Only Input

Image Input

Video Input

Instruct (or Non-Thinking) Mode

[!Important] Qwen3.5 does not officially support the soft switch of Qwen3, i.e., /think and /nothink.

[!Note] If you are using APIs from Alibaba Cloud Model Studio, in addition to changing model, please use "enable_thinking": False instead of "chat_template_kwargs": {"enable_thinking": False}.

Agentic Usage

Qwen-Agent

Qwen Code

Processing Ultra-Long Texts

• Modifying the model configuration file: In the config.json file, change the rope_parameters fields in text_config to:
  json

  1{ 2 "mrope_interleaved": true, 3 "mrope_section": [ 4 11, 5 11, 6 10 7 ], 8 "rope_type": "yarn", 9 "rope_theta": 10000000, 10 "partial_rotary_factor": 0.25, 11 "factor": 4.0, 12 "original_max_position_embeddings": 262144, 13}
• Passing command line arguments:
  For vllm, you can use
  VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1010000
  For sglang and ktransformers, you can use
  SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --context-length 1010000

[!NOTE] All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, potentially impacting performance on shorter texts. We advise modifying the rope_parameters configuration only when processing long contexts is required. It is also recommended to modify the factor as needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to set factor as 2.0.

Best Practices

1. Sampling Parameters:
   • We suggest using the following sets of sampling parameters depending on the mode and task type:
     • Thinking mode for general tasks:
       temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
     • Thinking mode for precise coding tasks (e.g., WebDev):
       temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
     • Instruct (or non-thinking) mode for general tasks:
       temperature=0.7, top_p=0.8, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
     • Instruct (or non-thinking) mode for reasoning tasks:
       temperature=1.0, top_p=1.0, top_k=40, min_p=0.0, presence_penalty=2.0, repetition_penalty=1.0
   • For supported frameworks, you can adjust the presence_penalty parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
2. Adequate Output Length: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 81,920 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
3. Standardize Output Format: We recommend using prompts to standardize model outputs when benchmarking.
   • Math Problems: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
   • Multiple-Choice Questions: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the answer field with only the choice letter, e.g., "answer": "C"."
4. No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.
5. Long Video Understanding: To optimize inference efficiency for plain text and images, the size parameter in the released video_preprocessor_config.json is conservatively configured. It is recommended to set the longest_edge parameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example,
   {"longest_edge": 469762048, "shortest_edge": 4096}
   Alternatively, override the default values via engine startup parameters. For implementation details, refer to: vLLM / SGLang.

Citation

The Window That Exploded

User

Assistant

1. Analyze the Request:
   • Length: Approximately 2000 words.
   • Style: Vivid, graphic horror.
   • POV: First Person ("I").
   • Character: Diana.
   • Setting: A skyscraper, specifically looking out a window on the 21st floor.
   • Inciting Incident/Imagery: The window explodes (or rather, the scene starts with the building swaying and the window exploding as a focal point).
   • Goal: Create an immersive, visceral experience.
2. Brainstorming & Outlining:
   • Opening: Needs to establish the swaying motion immediately. The wind, the tension before the explosion.
   • The Explosion: Not just glass shattering, but a release of pressure, a projectile of glass and perhaps something else (dust, blood, rain?). Let's make it about the sound and the sensation of pressure.
   • The Victim/Target: Is it me? Am I the one watching? Or is it me and a woman named Diana (since the prompt says POV character Diana)? Let's keep it consistent: Diana is the narrator.
   • The Horror: What happened to the people in the window? Or was it just the glass? Maybe the glass shattered because the air rushed out from a vacuum formed by someone dying, or simply because the building's integrity is failing. Let's go with the building failing, but hinting that we are next.
   • Sensory Details: Smell of ozone, taste of metal, vibration in teeth, cold sweat.
   • Structure:
     • Part 1: The Sway (0-500 words) - Building up anticipation.
     • Part 2: The Event (500-1000 words) - The explosion, the rush of wind, the debris.
     • Part 3: The Aftermath/Realization (1000-1500 words) - Dust settles, the realization of how close death came.
     • Part 4: The Descent/Conclusion (1500-2000 words) - Moving to the next floor or realizing the pattern continues.
3. Drafting - Section by Section:
   • Section 1: The Sway. Focus on the wind. It's unnatural. It's not a normal breeze. Diana is standing by the window. The 21st floor.
     • Keywords: Leaning, tectonic, simmering, metal bones, perspiration.
     • Sensory: Humming, vibrating floor.
     • Voice: Introspective, trembling.
   • Section 2: The Explosion.
     • Action: The window gives way.
     • Visuals: Mosaic of diamond dust. A vacuum sucking her in, or just blowing her hair. Let's have her recoil.
     • Sound: A cannon shot, a whistle.
     • Graphic Element: Blood? Maybe a drop of blood fell from a ceiling crack caused by the shockwave. Let's add that. The ceiling cracked.
     • The "Thing" in the Window: Was it just the glass? Or did someone inside scream before they died? Let's introduce the idea that the person inside the window was crushed or blown out. No, simpler: The window exploded inward, spraying Diana. Wait, if it explodes in front of her, it hits her face.
     • Revision: The window explodes inward, slamming against her. No, the prompt says "watch the window in front of me explode." So she stands watching it shatter outward, sending shards into the street, but she feels the air pressure. Or maybe the window explodes outward, and she has to dodge. Let's make it visceral. It's a storm of glass.
     • Refinement: To maximize horror, the window should explode, spraying glass everywhere. She is a fly on the wall.
   • Section 3: The Aftermath & Internal State.
     • Dust settles. The smell of broken silica.
     • Her hands are bleeding from a cut she didn't feel.
     • The sound of sirens far below.
     • The realization that the 21st floor isn't the problem; it's just the middle of a nightmare.
     • Is it just her, or everyone? A montage of chaos below.
     • The elevator shaft screaming.
   • Section 4: The Pattern.
     • She looks down. The other windows are darkening.
     • The wind dies, but the silence is loud.
     • The sun sets abruptly because the sky was torn open (maybe?).
     • She steps back. The floor groans.
     • Closing thought: How do you survive one night of this?
4. Writing - Focusing on Flow and Word Count:
   • Need to ensure the pacing slows down to let the details sink in.
   • Check adjectives and verbs. Avoid generic words (look, see) and use specific ones (gaze, witness).
   • Self-Correction during drafting: I need to make sure the "graphic" part lands. Not just blood, but the feeling of the environment attacking her.
   • Word Count Check: If it's running short, expand on the micro-movements of her body. The tremble of a finger, the dilation of a pupil.
5. Refining the Narrative Arc:
   • Beginning: Diana on the 21st floor. Wind rising.
   • Middle: The event. The explosion.
   • End: The lingering dread.
   • Key motif: The color blue (steel, sky, glass).
6. Drafting Phase (Mental or Scratchpad):
   • Start with the swaying. "The skyscraper swayed."
   • Build the pressure.
   • The explosion: "BANG."
   • The detail: A single drop of blood on her lip from the shock.
   • The view down: The city looking up like a spiderweb of light.
   • Ending: She realizes she is now trapped by the storm.
   (Self-Correction during mental drafting): Wait, the prompt says "The skyscraper sways, as I watch the window in front of me on the 21st floor explode..." Okay, that's the core image. I need to build the scene around that specific moment.
   Let's flesh out the "Graphic Horror" aspect. It shouldn't just be the window. It should be the air, the light, the dust. Maybe the person behind the glass died before they died. Let's have the explosion reveal a body, or the remnants of a body, inside the window. No, keep it focused on Diana's experience. The horror is existential. Let's introduce a smell. Burnt hair? Plastic? Yes.
   Let's count the rhythm. Short sentences for impact. Long sentences for flow. Start drafting.
   Reviewing against constraints:
   • 2000 words approx.
   • Vivid, graphic.
   • First Person.
   • Name: Diana.
   • POV: Standing at window.
   (Final Plan)
   1. Establish the wind and the structural integrity failing.
   2. The countdown in her mind as the window holds.
   3. The explosion: kinetic energy, sound, glass.
   4. The debris on the floor.
   5. The realization that she is the next victim or the observer of the victim?
   6. Moving to the balcony.
   7. A sense of the city waking up from a deep sleep into a fever dream.
   Let's go.