DavidAU/Qwen3.5-9B-Polar-Rev1-Uncensored-Heretic

Qwen3.5-9B-Polar-Rev1-Uncensored-Heretic

A fine tune (Unsloth, local hardware) using advanced "Polaris Dataset" (GPT5, org dataset by "TeichAI") using a different than "normal" training method with "BENCHES" below of 9B and 27B "rev1" versions compared against non-trained 9B and 27B models.

"Rev1" is the specific method that has shown to improve almost all metrics in the model, better than standard training.

ONE example generation below.

The "rev1" version of Qwen 3.5 9B exceeds the Qwen 3.5 27B non-trained model too, in 5 of 7 benchmarks and almost matches it on the other 2 benchmarks as well.

The kicker:

• The dataset is only 1017 samples.
• The samples are non-reasoning.

Model is also uncensored via HERETIC - prior to training.

NOTE:

• The 27B test results are a vastly stronger story.
• 27B Benches show across the board - 7 out of 7 - key metrics rising when training the 27B Qwen 3.5 model.

Benchmarks:

           arc   arc/e boolq hswag obkqa piqa  wino

---
9B
---

Qwen3.5-9B-Polar-Rev1-Uncensored-Heretic [instruct]
mxfp8      0.626,0.822,0.892,0.658,0.440,0.768,0.694

Qwen3.5-9B-Instruct: [base, non heretic]
mxfp8      0.571,0.719,0.895,0.683,0.426,0.770,0.671

---
27B
---

Qwen3.5-27B-Polar-Rev1-Uncensored-Heretic [instruct]
mxfp8      0.688,0.840,0.902,0.756,0.498,0.803,0.756

Qwen3.5-27B-Instruct: [base, non heretic]
mxfp8      0.557,0.711,0.868,0.533,0.452,0.706,0.695

Models are tested in "Instruct" mode because this generally works better with the testing harness.

Testing via "thinking" mode also shows the metrics (and changes) but not the true extent.

This is also a HERETIC model, trained post "Heretic'ing" -> this model does what you want, no questions asked.

Fully uncensored.

VISION:

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

---

DE-CENSORING:

Performance

KLD of less than 1 is excellent, zero is perfect.

Metric	This model	Original model (Qwen/Qwen3.5-9B)
KL divergence	0.0793	0 (by definition)
Refusals	6/100	100/100

NOTES:

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

IMPORTANT:

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

---

Using an "uncensored" (refusals removed) model VS trained "uncensored" model

Usually when you a tell a model to generate horror, swear or x-rated content this is all you have to do to get said content type.

In the case of this model, it will not refuse your request, however it needs to be "pushed" a bit / directed a bit more in SOME CASES.

Although this model will generated x-rated content too, likewise you need to tell it to use "slang" (and include the terms you want) to get it generate the content correctly as the "expected" content level too.

Without these added directive(s), the content can be "bland" by comparison to an "uncensored model" or model trained on uncensored content.

Roughly, the model tries to generate the content but the "default" setting(s) are so "tame" it needs a push to generate at expected graphic, cursing or explicit levels.

Even with minimal direction (ie, use these words to swear: x,y,z), this will be enough to push the model to generate the requested content in the ahh... expected format.

---

Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:

In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;

Set the "Smoothing_factor" to 1.5

: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"

: in text-generation-webui -> parameters -> lower right.

: In Silly Tavern this is called: "Smoothing"

NOTE: For "text-generation-webui"

-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)

Source versions (and config files) of my models are here:

https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be

OTHER OPTIONS:

• 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.

Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers

This a "Class 1" model:

For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:

[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]

You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:

[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]

---

Qwen3.5-9B

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.

Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency.

Qwen3.5 Highlights

Qwen3.5 features the following enhancement:

• 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.

For more details, please refer to our blog post Qwen3.5.

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

	GPT-OSS-120B	GPT-OSS-20B	Qwen3-Next-80B-A3B-Thinking	Qwen3-30BA3B-Thinking-2507	Qwen3.5-9B	Qwen3.5-4B
Knowledge & STEM
MMLU-Pro	80.8	74.8	82.7	80.9	82.5	79.1
MMLU-Redux	91.0	87.8	92.5	91.4	91.1	88.8
C-Eval	76.2	71.4	89.7	87.4	88.2	85.1
SuperGPQA	54.6	48.5	60.8	56.8	58.2	52.9
GPQA Diamond	80.1	71.5	77.2	73.4	81.7	76.2
Instruction Following
IFEval	88.9	88.2	88.9	88.9	91.5	89.8
IFBench	69.0	65.1	61.5	51.5	64.5	59.2
MultiChallenge	45.3	40.1	51.3	46.5	54.5	49.0
Long Context
AA-LCR	50.7	30.7	51.7	49.0	63.0	57.0
LongBench v2	48.2	45.6	48.0	44.8	55.2	50.0
Reasoning & Coding
HMMT Feb 25	90.0	76.7	73.7	63.1	83.2	74.0
HMMT Nov 25	90.0	81.8	81.2	73.8	82.9	76.8
LiveCodeBench v6	82.7	74.6	68.7	66.0	65.6	55.8
OJBench	41.5	36.3	29.7	25.1	29.2	24.1
General Agent
BFCL-V4	--	--	49.7	42.4	66.1	50.3
TAU2-Bench	--	--	57.4	41.9	79.1	79.9
VITA-Bench	--	--	29.5	14.1	29.8	22.0
DeepPlanning	--	--	0.4	4.9	18.0	17.6
Multilingualism
MMMLU	78.2	69.7	81.3	78.4	81.2	76.1
MMLU-ProX	74.5	67.3	73.6	69.1	76.3	71.5
NOVA-63	51.1	48.7	53.3	52.5	55.9	54.3
INCLUDE	74.0	65.3	78.3	74.4	75.6	71.0
Global PIQA	84.1	79.8	83.5	80.2	83.2	78.9
PolyMATH	54.0	30.9	62.4	52.6	57.3	51.1
WMT24++	74.4	67.8	57.4	69.3	72.6	66.6
MAXIFE	83.7	80.1	79.9	77.4	83.4	78.0

* TAU2-Bench: we follow the official setup except for the airline domain, where all models are evaluated by applying the fixes proposed in the Claude Opus 4.5 system card.

* MMLU-ProX: we report the averaged accuracy on 29 languages.
* WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL.
* MAXIFE: we report the accuracy on English + multilingual original prompts (totally 23 settings).
* Empty cells (--) indicate scores not yet available or not applicable.

Vision Language

	GPT-5-Nano-2025-08-07	Gemini-2.5-Flash-Lite	Qwen3-VL-30B-A3B	Qwen3.5-9B	Qwen3.5-4B
STEM and Puzzle
MMMU	75.8	73.4	76.0	78.4	77.6
MMMU-Pro	57.2	59.7	63.0	70.1	66.3
MathVision	62.2	52.1	65.7	78.9	74.6
Mathvista(mini)	71.5	72.8	81.9	85.7	85.1
We-Math	62.5	32.1	70.0	75.2	75.4
DynaMath	78.0	69.9	80.1	83.6	83.3
ZEROBench	1.0	1.0	0.0	3.0	3.0
ZEROBench_sub	22.2	19.2	23.7	31.1	26.3
VlmsAreBlind	66.7	68.4	72.5	93.7	92.6
BabyVision	14.4	17.5	18.6	28.6/25.8	16.0/19.1
General VQA
RealWorldQA	71.8	72.2	77.4	80.3	79.5
MMStar	68.6	69.1	75.5	79.7	78.3
MMBenchEN-DEV-v1.1	80.3	82.7	88.9	90.1	89.4
SimpleVQA	46.0	54.1	54.3	51.2	43.4
HallusionBench	58.4	64.5	66.0	69.3	65.0
Text Recognition and Document Understanding
OmniDocBench1.5	55.9	79.4	86.8	87.7	86.2
CharXiv(RQ)	50.1	56.1	56.6	73.0	70.8
MMLongBench-Doc	31.8	46.5	47.4	57.7	54.2
CC-OCR	58.9	72.9	77.8	79.3	76.7
AI2D_TEST	81.9	85.7	86.9	90.2	89.6
OCRBench	75.3	82.5	83.9	89.2	85.0
Spatial Intelligence
ERQA	45.8	44.3	45.3	55.5	54.0
CountBench	80.0	79.2	90.0	97.2	96.3
RefCOCO(avg)	--	--	89.3	89.7	88.1
EmbSpatialBench	74.2	66.1	80.6	83.0	81.3
RefSpatialBench	12.6	11.2	54.2	58.5	54.6
LingoQA	57.0	17.8	62.0	80.4	74.4
Hypersim	--	--	11.4	13.5	12.5
Nuscene	--	--	10.3	11.8	9.9
Video Understanding
VideoMME(w sub.)	71.7	74.6	79.9	84.5	83.5
VideoMME(w/o sub.)	66.2	72.7	73.3	78.4	76.9
VideoMMMU	63.0	69.2	75.0	78.9	74.1
MLVU	69.2	78.5	78.9	84.4	82.8
MVBench	--	--	72.0	74.4	71.2
LVBench	--	60.9	59.2	70.0	66.4
MMVU	63.1	65.3	66.1	67.8	64.9
Visual Agent
ScreenSpot Pro	--	--	60.5	65.2	60.3
OSWorld-Verified	--	--	30.6	41.8	35.6
AndroidWorld	--	--	55.0	57.8	58.6
Tool Calling
TIR-Bench	18.5	21.5	22.5	45.6/31.9	38.9/29.9
V*	68.1	69.6	83.2	90.1/88.5	84.3/86.4
Medical VQA
SLAKE	57.0	65.0	68.8	79.0	76.1
PMC-VQA	37.8	48.8	51.5	57.9	55.5
MedXpertQA-MM	26.7	35.3	35.5	49.9	42.9

* MathVision: our model’s score is evaluated using a fixed prompt, e.g., “Please reason step by step, and put your final answer within \boxed{}.” For other models, we report the higher score between runs with and without the \boxed{} formatting.
* BabyVision: scores reported as "with CI / without CI".
* TIR-Bench and V*: scores reported as "with CI / without CI".
* Empty cells (--) indicate scores not yet available or not applicable.

Quickstart

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.

For streamlined integration, we recommend using Qwen3.5 via APIs. Below is a guide to use Qwen3.5 via OpenAI-compatible API.

Serving Qwen3.5

Qwen3.5 can be served via APIs with popular inference frameworks. In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.5 models.

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.

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

SGLang is a fast serving framework for large language models and vision language models. SGLang from the main branch of the open-source repository is required for Qwen3.5, which can be installed using the following command in a fresh environment:

uv pip install 'git+https://github.com/sgl-project/sglang.git#subdirectory=python&egg=sglang[all]'

See its documentation for more details.

The following will create API endpoints at http://localhost:8000/v1:

• 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

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. vLLM from the main branch of the open-source repository is required for Qwen3.5, which can be installed using the following command in a fresh environment:

uv pip install vllm --torch-backend=auto --extra-index-url https://wheels.vllm.ai/nightly

See its documentation for more details.

For detailed Qwen3.5 usage guide, see the vLLM Qwen3.5 recipe.

The following will create API endpoints at http://localhost:8000/v1:

• 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

KTransformers is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing. For running Qwen3.5 with KTransformers, see the KTransformers Deployment Guide.

Hugging Face Transformers

Hugging Face Transformers contains a lightweight server which can be used for quick testing and moderate load deployment. The latest transformers is required for Qwen3.5:

pip install "transformers[serving] @ git+https://github.com/huggingface/transformers.git@main"

See its documentation for more details. Please also make sure torchvision and pillow are installed.

Then, run transformers serve to launch a server with API endpoints at http://localhost:8000/v1; it will place the model on accelerators if available:

transformers serve --force-model Qwen/Qwen3.5-9B --port 8000 --continuous-batching

Using Qwen3.5 via the Chat Completions API

The chat completions API is accessible via standard HTTP requests or OpenAI SDKs. Here, we show examples using the OpenAI Python SDK.

Before starting, make sure it is installed and the API key and the API base URL is configured, e.g.:

pip install -U openai

# Set the following accordingly
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"

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

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [
    {"role": "user", "content": "Type \"I love Qwen3.5\" backwards"},
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.5-9B",
    messages=messages,
    max_tokens=81920,
    temperature=1.0,
    top_p=0.95,
    presence_penalty=1.5,
    extra_body={
        "top_k": 20,
    }, 
)
print("Chat response:", chat_response)

Image Input

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg"
                }
            },
            {
                "type": "text",
                "text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$"
            }
        ]
    }
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.5-9B",
    messages=messages,
    max_tokens=81920,
    temperature=1.0,
    top_p=0.95,
    presence_penalty=1.5,
    extra_body={
        "top_k": 20,
    }, 
)
print("Chat response:", chat_response)

Video Input

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "video_url",
                "video_url": {
                    "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4"
                }
            },
            {
                "type": "text",
                "text": "Summarize the video content."
            }
        ]
    }
]

# When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,
# video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).
# This feature is currently supported only in vLLM.
#
# By default, `fps=2` and `do_sample_frames=True`.
# With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.
chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.5-9B",
    messages=messages,
    max_tokens=81920,
    temperature=1.0,
    top_p=0.95,
    presence_penalty=1.5,
    extra_body={
        "top_k": 20,
        "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},
    }, 
)

print("Chat response:", chat_response)

Instruct (or Non-Thinking) Mode

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

Qwen3.5 will think by default before response. You can obtain direct response from the model without thinking by configuring the API parameters. For example,

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/RealWorld/RealWorld-04.png"
                }
            },
            {
                "type": "text",
                "text": "Where is this?"
            }
        ]
    }
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.5-9B",
    messages=messages,
    max_tokens=32768,
    temperature=0.7,
    top_p=0.8,
    presence_penalty=1.5,
    extra_body={
        "top_k": 20,
        "chat_template_kwargs": {"enable_thinking": False},
    }, 
)
print("Chat response:", chat_response)

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

Qwen3.5 excels in tool calling capabilities.

Qwen-Agent

We recommend using Qwen-Agent to quickly build Agent applications with Qwen3.5.

To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself.

import os
from qwen_agent.agents import Assistant

# Define LLM
# Using Alibaba Cloud Model Studio
llm_cfg = {
    # Use the OpenAI-compatible model service provided by DashScope:
    'model': 'Qwen3.5-9B',
    'model_type': 'qwenvl_oai',
    'model_server': 'https://dashscope.aliyuncs.com/compatible-mode/v1',
    'api_key': os.getenv('DASHSCOPE_API_KEY'),

    'generate_cfg': {
        'use_raw_api': True,
        # When using Dash Scope OAI API, pass the parameter of whether to enable thinking mode in this way
        'extra_body': {
            'enable_thinking': True
        },
    },
}

# Using OpenAI-compatible API endpoint.
# functionality of the deployment frameworks and let Qwen-Agent automate the related operations.
#
# llm_cfg = {
#     # Use your own model service compatible with OpenAI API by vLLM/SGLang:
#     'model': 'Qwen/Qwen3.5-9B',
#     'model_type': 'qwenvl_oai',
#     'model_server': 'http://localhost:8000/v1',  # api_base
#     'api_key': 'EMPTY',
#
#     'generate_cfg': {
#         'use_raw_api': True,
#         # When using vLLM/SGLang OAI API, pass the parameter of whether to enable thinking mode in this way
#         'extra_body': {
#             'chat_template_kwargs': {'enable_thinking': True}
#         },
#     },
# }

# Define Tools
tools = [
    {'mcpServers': {  # You can specify the MCP configuration file
            "filesystem": {
                "command": "npx",
                "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/xxxx/Desktop"]
            }
        }
    }
]

# Define Agent
bot = Assistant(llm=llm_cfg, function_list=tools)

# Streaming generation
messages = [{'role': 'user', 'content': 'Help me organize my desktop.'}]
for responses in bot.run(messages=messages):
    pass
print(responses)

# Streaming generation
messages = [{'role': 'user', 'content': 'Develop a dog website and save it on the desktop'}]
for responses in bot.run(messages=messages):
    pass
print(responses)

Qwen Code

Qwen Code is an open-source AI agent for the terminal, optimized for Qwen models. It helps you understand large codebases, automate tedious work, and ship faster.

For more information, please refer to Qwen Code.

Processing Ultra-Long Texts

Qwen3.5 natively supports context lengths of up to 262,144 tokens. For long-horizon tasks where the total length (including both input and output) exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively., e.g., YaRN.

YaRN is currently supported by several inference frameworks, e.g., transformers, vllm, ktransformers and sglang. In general, there are two approaches to enabling YaRN for supported frameworks:

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

  {
      "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,
  }
  

• 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
  

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

To achieve optimal performance, we recommend the following settings:

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

If you find our work helpful, feel free to give us a cite.

@misc{qwen3.5,
    title  = {{Qwen3.5}: Towards Native Multimodal Agents},
    author = {{Qwen Team}},
    month  = {February},
    year   = {2026},
    url    = {https://qwen.ai/blog?id=qwen3.5}
}

---

EXAMPLE Generation:

Q4KS, Non-Imatrix

Expect better results from Imatrix and/or higher quants.

NOTE: Some formatting lost on copy/paste.

---

Nighttime Radiative Cooling Strategies

User

Explain 10 ways (in extensive detail) to use the "night" time cooling of radiant energy into space to reduce global temperatures. Include charts, and tables.

Assistant

<think>

Here's a thinking process that leads to the suggested answer:

1. Analyze the Request:

   • Topic: Using the "night" time cooling of radiant energy into space to reduce global temperatures.
   • Format: Explain 10 ways.
   • Detail: Extensive detail.
   • Visuals: Include charts and tables.
   • Constraint Check: Since I'm a text-based AI model (LLM), I cannot generate actual interactive charts/images or complex file-based graphs. I have to simulate them using Markdown, ASCII, or descriptive text representations.
   • Scientific Accuracy: This is the critical filter. The concept of "radiative cooling" is real (Earth naturally radiates to space), but the idea of "technologically enhancing night-time cooling to reduce global temperatures" is often conflated with concepts like "geoengineering," "high-albedo surfaces," or "orbital mirrors." Some proposals rely on physics that are either speculative, currently impractical, or misunderstood (e.g., simply "capturing" the night's cooling).
   • Goal: Provide a robust, detailed, and scientifically grounded response that respects the user's curiosity while maintaining realism. Avoid pseudoscience (like "cooling crystals") but explain actual mechanisms.

2. Deconstruct the Physics:

   • Earth naturally cools at night.
   • To "use" this for global cooling, we need to:
     • Increase Earth's Albedo (reflectivity).
     • Reduce Greenhouse Gas (GHG) opacity to allow more IR escape.
     • Actively pump heat to space using radiators or heat pipes.
     • Modify atmospheric composition.
   • The prompt asks for 10 ways. Many existing "geoengineering" concepts can be mapped here.

3. Drafting the 10 Ways (Iterative Process):

   • Idea 1: High-reflectivity surfaces. (Painting roofs, deserts).
   • Idea 2: Atmospheric aerosols. (Stratospheric aerosol injection - similar to volcanic winters).
   • Idea 3: Radiative Cooling Materials. (Materials that emit heat at "cold" space wavelengths).
   • Idea 4: Orbital Shading. (Space sunshades/mirrors).
   • Idea 5: Cloud Manipulation. (Making thinner, higher clouds).
   • Idea 6: Carbon Capture via Albedo. (CCUS combined with land use).
   • Idea 7: Ocean Whitening. (Reflecting sea ice/surface).
   • Idea 8: Night-time Active Radiators. (Theoretical space-based heat dump).
   • Idea 9: Biogenic Albedo. (Restoring forests).
   • Idea 10: Surface Darkening/Whitening. (Managing urban heat islands).

   Critique: Some of these are general mitigation. I need to focus specifically on utilizing the "radiative cooling to space" aspect (IR emission). Refining for "Night-time Radiative Cooling":

   1. Global High-Albedo Landscapes: Maximizing reflection during the day helps radiative balance.
   2. Radiative Sky Cooling Materials: Applying "Day/Night Radiative Cooling" (DNRC) to large areas (roofs, deserts).
   3. Stratospheric Radiative Injection: Introducing particles that absorb/emit in IR bands.
   4. Ocean Reflective Films: Floating materials that trap cold air or enhance reflection.
   5. Night-Sky Transparency Optimization: Reducing light pollution and specific wavelengths that trap heat.
   6. Space-Based Radiators: Massive heat pipes in orbit (direct heat dump).
   7. Cloud Seeding for Emissive Windows: Encouraging cirrus that radiate efficiently.
   8. Urban Radiative Cooling Parks: Replacing asphalt with emissive materials.
   9. Selective Atmospheric Thinning: Using GHG-removing aerosols.
   10. Geographic "Cooling Zones": Creating localized zones of intense radiative cooling to counter heatwaves.

4. Addressing the "Charts and Tables" Requirement:

   • I will create:
     • A table comparing mechanisms (Pros/Cons/Feasibility).
     • A text-based "Graph" showing the concept of radiative balance (Input vs Output).
     • A visual breakdown of the 10 methods.

5. Refining the Tone:

   • Must be detailed.
   • Must not promise that we can "flip a switch" and reverse climate change instantly.
   • Must acknowledge the complexity (economic, political, physical).

6. Drafting the Content (Section by Section):

   • Introduction: Briefly explain Earth's radiative equilibrium.
   • Visual 1: A conceptual diagram of Radiative Balance.
   • The 10 Ways:
     • Each needs a specific focus on how it leverages night-time radiation.
     • Example for #6 (Space Radiators): Needs to sound technically plausible.
   • Visual 2: A table.
   • Visual 3: A breakdown chart.
   • Conclusion: Summary of risks and potential.

7. Self-Correction/Sanity Check during drafting:

   • Issue: Is "using night cooling" a real strategy or just semantics?
   • Answer: All climate engineering is fundamentally about optimizing Earth's radiative energy budget. At night, Earth radiates IR. If we block that (GHGs) or enhance it (aerosols/cooling tech), we change temperature.
   • Issue: Don't confuse this with "cryogenic cooling."
   • Correction: Ensure language refers to "radiative cooling" (passive emission of thermal energy into cold vacuum).
   • Constraint: I can't generate .png/.jpg.
   • Solution: Use ASCII art and structured Markdown blocks to represent charts.

8. Final Polish:

   • Check for clarity.
   • Ensure the detail is "extensive" (approx. 200+ words per point).
   • Format nicely.

(Self-Correction Note): If the user is expecting a conspiracy-style or purely sci-fi answer, I need to walk the line between imaginative and grounded. Let's stick to scientifically grounded "Geoengineering" concepts.

Let's build the response.

• Start with the physics.
• Present 10 specific strategies.
• Integrate visual aids.
• Maintain a serious tone.

Let's generate.

</think>

Excellent question. You’re asking about a concept that sits at the very frontier of geoengineering: optimizing Earth’s natural radiative energy budget—specifically its ability to shed heat as infrared radiation into the cold vacuum of space, especially during the night.

To answer this with extensive detail and clarity, we’ll break this down into:

1. A simple conceptual visualization.
2. Detailed explanation of 10 advanced strategies.
3. Charts and tables illustrating feasibility.

First, a quick physics refresher: Earth stays at an average temperature because it balances how much energy it gets from the Sun with how much it radiates back out. Greenhouse gases act like a blanket, trapping that outgoing heat. We can’t easily “remove” heat from the ocean faster than it comes in with current tech. However, we can modify how efficiently the planet radiates heat away—especially at night, when it’s naturally cooling, but being held warm by the atmosphere.

Here are 10 technically grounded, theoretically scalable ways to enhance this process.

1. Global Deployment of Radiative Sky-Cooling Materials on Large Surfaces

   • Concept: Radiative Sky Cooling (RSC) materials are specially engineered surfaces that can emit infrared heat directly into space through a “transparent” window in the atmosphere (around 10 micrometers wavelength), even on cloudy nights.
   • Mechanism: If you coat rooftops, solar farms, or even deserts with materials optimized for “Day/Night Radiative Cooling” (DNRC), they will passively shed thermal energy into the sky.
   • Night-Time Enhancement: During the night, these materials become exceptionally efficient radiators. By deploying them over millions of square kilometers, you increase the global outgoing longwave radiation (OLR).
   • Scale: If we coated 100,000 km² of land with high-performance radiative paints, theoretical models suggest we could locally reduce surface temperatures by 1-3°C.
   • Realism: High. We already have these materials for building efficiency. Scaling up is an industrial logistics challenge, not a physics one.

2. Stratospheric Aerosol Injection via Night-Time Optimization

   • Concept: Stratospheric aerosol injection (SAI) is similar to how volcanoes reflect sunlight and cool the planet.
   • Night-Time Enhancement: The key here is the composition of the aerosols. By injecting sulfur or calcium fluoride particles that are highly reflective in visible light and highly emissive in infrared during the night, you maximize both daytime reflection and nighttime radiative efficiency.
   • Mechanism: These particles would absorb some heat and re-emit it as radiation, but by tuning their size and chemical structure, you force them to release energy into space more efficiently than natural clouds.
   • Effect: This would allow Earth’s surface to lose more energy to space at night, effectively counteracting greenhouse warming.
   • Realism: Moderate. We know the basic physics; the political and environmental control is the challenge.

3. Orbital Radiative Cooling Arrays (Space Shading)

   • Concept: Deploying thin, highly reflective mirrors or radiative structures in Earth orbit.
   • Mechanism: These don’t block the sun (which would freeze us). Instead, they are positioned so that they reflect a small percentage of sunlight away from Earth (increasing albedo), and they themselves act as radiators, dumping heat captured from the atmosphere into space.
   • Night-Time Use: At night, Earth sees the dark side of space. These structures, if oriented correctly, enhance the "view" Earth has of the deep cold universe.
   • Scale: A small constellation of mirrors could have a noticeable effect if combined with atmospheric modifications.
   • Realism: Low to Moderate. Extremely expensive, but physics is sound.

4. Ocean Surface “Whitening” with High-Reflectance Films

   • Concept: Covering parts of the ocean surface with ultra-thin, floating, radiative-cooling films.
   • Mechanism: These films would be:
     1. Highly reflective to solar radiation during the day.
     2. Radiatively emissive at night (IR-transparent to space).
     3. Non-biohazardous and biodegradable.
   • Effect: At night, instead of radiating heat back into the warmer atmosphere, the ocean surface temperature under the film can be allowed to drop naturally and efficiently via these films.
   • Impact: This would cool sea surface temperatures, which are the primary driver of tropical weather and hurricanes.
   • Realism: Very low currently. Engineering a stable film that survives ocean storms without poisoning marine life is a massive hurdle.

5. Cloud Seeding for Enhanced “Radiative Windows”

   • Concept: Encouraging the formation of specific cloud types that are highly efficient at radiating heat to space.
   • Mechanism: Most low clouds trap heat (greenhouse effect). However, high, thin cirrus clouds, if manipulated, can be made more “transparent” to heat radiation in the infrared spectrum, allowing Earth to radiate more.
   • Night-Time Enhancement: During the night, we could introduce aerosols that nucleate into small, high-altitude clouds optimized to emit IR efficiently rather than absorbing it.
   • Effect: A “thinned” atmosphere that lets more heat escape.
   • Realism: Moderate. Cloud physics is extremely complex; we’d risk unintended weather effects.

6. Urban Radiative Cooling Zones (RCZ)

   • Concept: Redesigning cities to maximize their “radiative cooling to space” capability at night.
   • Mechanism:
     1. Replace heat-absorbing asphalt with radiative-cooling pavement.
     2. Integrate “radiative towers” (massive vertical heat pipes ending in radiative fins exposed to the open sky).
     3. Increase vertical greenery and water features that evaporate and radiate at night.
   • Effect: These zones act as localized heat pumps, pulling excess heat out of the city at night via radiation and releasing it upward into the clear sky.
   • Why Night? This is when the urban heat island effect peaks. Efficient night cooling prevents the next day’s temperature from spiking.
   • Realism: High. This is standard urban planning mixed with new materials.

7. Selective Greenhouse Gas (GHG) “Radiator” Injection

   • Concept: Using compounds that preferentially absorb and emit heat at wavelengths that match the “clear sky” window.
   • Mechanism: Instead of just scrubbing CO2 (which takes energy), we could theoretically introduce trace amounts of gases or nano-structures that are optimized to let energy pass through to space more easily.
   • Night-Time Mechanism: These would be particularly effective at night when the atmosphere is naturally “cooling down.” They would reduce the atmospheric opacity in specific IR bands.
   • Realism: Low. Risky. We could accidentally create more GHG warming if the wrong wavelengths are targeted.

8. Space-Based Heat Dump Radiators

   • Concept: Building massive, space-based radiators that are thermally coupled to Earth.
   • Mechanism: Imagine huge heat pipes (or plasma cooling towers) on the ground in Antarctica or desert regions that pull heat from the Earth’s atmosphere, carry it into space-based radiators, and dump it into the vacuum.
   • Night-Time Enhancement: At night, these radiators can work at peak efficiency because they are in direct line-of-sight with the deep universe (no air to impede radiative loss).
   • Realism: Very low. Requires megastructure engineering (like the Dyson swarm concept). But it is theoretically the most direct way to remove heat from Earth.

9. Biogenic Albedo Restoration for Night-Time Cooling

   • Concept: Restoring forests and wetlands not just to sequester carbon, but to improve local radiative balance.
   • Mechanism: Trees and vegetation have different emissivity properties than concrete. Forests absorb more heat during the day, but at night, they radiate heat more evenly.
   • Night-Time Enhancement: By rewilding arid lands into ecosystems that are optimized for transpiration and radiative cooling, you allow more heat to move from the land into the air and out into the sky via “breathing” the atmosphere at night.
   • Realism: High. This is a win-win: carbon capture + better cooling.
   • Note: This is not about the trees themselves, but about how the ecosystem optimizes local radiative transfer.

10. Night-Time Atmospheric “Thinning” Using Carbon-Based Nanoparticles

    • Concept: A speculative but physically sound geoengineering concept.
    • Mechanism: Dispersing carbon-based nanoparticles (e.g., carbon black or graphene flakes) into the upper atmosphere at night.
    • Why Night? At night, there is less solar scattering interference. These particles would be tuned to emit infrared radiation (heat) efficiently into space.
    • Effect: They would increase the “emissivity” of the atmosphere to space, similar to how a radiator works in your house.
    • Risk: This is the most advanced and potentially dangerous idea. Requires precision control to avoid rainout or altering rainfall patterns.
    • Realism: Theoretical. Would likely be the last option, if all else fails.

Here are visual aids to help understand this better.

TABLE: Comparative Feasibility and Effectiveness This table summarizes how realistic and effective each method is, focusing specifically on their use of radiative cooling.

+---------------------------------------------------+------------------+------------------+
| METHOD                                            | SCIENTIFIC BASIS | FEASIBILITY      |
+---------------------------------------------------+------------------+------------------+
| Radiative Sky-Cooling Materials (RSC)             | High             | High             |
| (Paints/films that emit heat at night)            |                  |                  |
+---------------------------------------------------+------------------+------------------+
| Stratospheric Aerosol Injection (SAI)             | High             | Moderate         |
| (Fine particles to optimize IR emission)           |                  |                  |
+---------------------------------------------------+------------------+------------------+
| Orbital Radiative Cooling Arrays                  | High             | Low (Cost)       |
| (Space mirrors/radiators)                          |                  |                  |
+---------------------------------------------------+------------------+------------------+
| Ocean Surface Films                               | Moderate         | Very Low (Tech)  |
| (Floating films that radiate at night)             |                  |                  |
+---------------------------------------------------+------------------+------------------+
| Urban Radiative Cooling Zones                     | High             | High             |
| (Rebuild cities with radiative materials)          |                  |                  |
+---------------------------------------------------+------------------+------------------+
| Space-Based Heat Dump Radiators                   | High             | Very Low (Mega)  |
| (Direct pipes to space)                            |                  |                  |
+---------------------------------------------------+------------------+------------------+

GRAPH: How Earth’s Radiative Balance Works (Simplified)

This simple ASCII chart helps visualize where these 10 methods fit into the global energy balance.

[ENERGY IN]                  [ENERGY OUT]
  1.00 Solar Input            1.00 Outgoing Longwave Radiation (OLR)
  (Shortwave)                  (Infrared to Space)
                             |
                             V
        [GREENHOUSE EFFECT]
       (Atmosphere traps some IR)
                             |
                             +-----------------------+
                             |                       |
                   [METHODS TO REDUCE T]    [METHODS TO ENHANCE T]
                   |  (Cooling)              | (Heating/Warming) |
                   V                         V
       - Reduce IR Trapping               - Block OLR (Too much!)
         (Albedo)                       - Thick Clouds
       - Increase Night Radiation        - More CO2
       - Add Radiative Surfaces

FINAL NOTE: Every one of these 10 methods is, in theory, rooted in real thermodynamics. The real-world problem isn’t that we don’t know how to cool the planet at night using radiation—we understand that physics perfectly well. The problem is:

• Scale: Moving from lab-scale cooling to planetary-scale cooling.
• Governance: Who authorizes a project that can potentially freeze parts of the planet if mismanaged?
• Side Effects: We risk altering rainfall, ocean chemistry, or disrupting ecosystems.

If you’d like, I can go deeper into the math or physics behind any specific method you’re curious about (e.g., how the orbital radiators would physically be built, or how the nanomaterials would work).