Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking-NEO-CODE-Di-IMatrix-MAX-GGUF
DavidAU's GGUF imatrix quantization of a 40B Qwen3.6 model fine-tuned for creative writing, coding, and roleplay without content filters.
Base model
Model Card
Ultimate NEO GGUF QUANTS: Custom built DUAL Imatrix NEO-CODER quants that exceed all other quants in terms of quality, stability, precision and long convo usage. IQ4_XS/NL regularly scores at 94% of full precision (bf16), Q6/Q8 at 97% and 98% of full precision (bf16).
WARNING: This model has character and intelligence. It will take no prisoners. It will give no quarter. Uncensored, Unfiltered and boldly confident. Not even remotely "SFW", if you ask it for NSFW content. And it is wickedly smart too - exceeding the base model in 6 out of 7 benchmarks.
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking-NEO-CODE-Di-IMatrix-MAX-GGUF
40 billion parameters (dense, not moe) expanded from 27B Qwen 3.6, then trained on Claude 4.6 Opus High Reasoning dataset via Unsloth on local hardware... but there is much more to the story - in comes DECKARD.
96 layers, 1275 Tensors. (50% more than base model of 27B)
Features variable length reasoning ; less complex = shorter, longer for more complex.
Model performance has increased dramatically. And it has character too.
A lot of character.
No censorship, no nanny. (via Heretic)
And it is very, very smart.
Fully uncensored first (via Heretic), then trained (via Unsloth) on "Deckard/PDK" internal datasets (5) (character, intelligence, depth, observation, and ah... point of view), THEN expanded to 40B parameters (room to think), and then trained (Unsloth again) with Claude 4.6 Opus Distill dataset (to shorten and improve reasoning, and stablize everything).
256K context.
"Thats no moon, thats a fully armed and operational Qwen-Station."
TWO example generations below [bottom of the page], more to come.
Brutal Honesty (on writing fiction, from this model: Q4KS, non imatrix):
Listen up, because I'm going to tell you something you probably don't want to hear: you're probably going to write a mediocre story on your own.Not because you're untalented—because writing fiction is hard as fuck. Even the greats needed editors, feedback, and someone to push back. That's where I come in, and I'm not just some AI tool you plug in like a microwave setting. I'm the collaborative partner you didn't know you needed until you've written 80,000 words of something that falls apart in the third act because you can't see the plot holes you've been digging since chapter two.
NEO-CODE-Di-IMatrix-MAX-GGUF Quants:
Quant "engineering" focused on balance and precision, vs raw power (which seemed in some cases to destabilize the model/quant).
In other words benchmarks / stats determined the best quants, not guesswork or one size fits all approach.
This was done to ensure long context, long/multi-convos, coding and math etc etc performed as close as possible to full precision model as well as one-shot, and standard prompting / problem solving.
TWO Imatrix datasets were used to do this by first getting "raw stats" on both, then merging them to get the best of each imatrix in one dataset then this was used to make the "NEO-CODE-Di-IMatrix-MAX" quants.
Additional tensor adjustments were also made, which were also measured (benched) and adjusted too.
How strong are they?
- IQ2_M -> 83-84% of BF16 full precision.
- IQ4XS -> 94% of BF16 full precision.
- Q8_0 HIGH -> 98.4% of BF16 full precision.
To see metrics [5 critical, and detailed] and stats on these engineered quants see these repos:
https://huggingface.co/DavidAU/Qwen3.6-27B-NEO-CODE-Di-IMatrix-MAX-GGUF
https://huggingface.co/DavidAU/Qwen3.6-27B-Heretic-Uncensored-FINETUNE-NEO-CODE-Di-IMatrix-MAX-GGUF
GGUF POWER UPS:
A radically stronger, more potent GGUF for all use cases.
Meets Unsloth quality, and exceeds it in some metrics (see below).
DETAILS:
- DI-MATRIX (duel imatrix) of NEO and NEO-CODE imatrix datasets (by DavidAU).
- All Unsloth tensor enhancements + additional enhancements CALIBRATED thru metrics testing.
- Every quant benchmarked against BF16/full precision model.
- There is a special Q8_0 quant, with BF16 components. Imatrix has no effect on Q8/BF16 tensors.
VISION:
- Vision (images) tested.
- You need an "mmproj" (just one) of these downloaded too, and placed in the same folder as the GGUF for images.
Qwen Model Settings (suggested):
- Thinking mode for general tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, 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: temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
- Context window min from 8k to 16k.
IMPORTANT: See also "CORE SETTINGS for 40B version" below.
Other Versions using Deckard/OPUS:
Qwen 3.5 40B Version: 181 likes and counting...
https://huggingface.co/DavidAU/Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking
GEMMA4 VERSIONS:
Examples and benchmarks.
GEMMA-4 31B Version, using the DECKARD datasets (5):
https://huggingface.co/DavidAU/gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking
GEMMA-4 19B-A4B (MOE) Versions, using the DECKARD datasets (5):
https://huggingface.co/DavidAU/gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-Thinking
https://huggingface.co/DavidAU/gemma-4-19B-A4B-it-The-DECKARD-Thinking
GEMMA-4 E4B (8B, moe like models), using the DECKARD datasets (5):
CORE SETTINGS for 40B version:
SETTINGS:
- min 8k to 16k context window.
- for creative rep pen of 1.05 to 1.1 WITH LOWER QUANTS.
- suggest temp .7 / rep pen 1 (off) for general usage.
- output generation can exceed 100k tokens.
- Suggest min quant of Q4KS (non imatrix) or IQ3_S (imatrix) or HIGHER.
- For toolcalls -> suggest Q5/Q6 min quants (as per Qwen guidence)
EXAMPLE SYSTEM PROMPTS:
The model does not need a system prompt, however if you want to enhance operation here are some samples.
#1 - All use cases.
Be vivid and precise.
#2 - Creative use cases:
Below is an instruction that describes a task. Ponder each user instruction carefully, and use your skillsets and critical instructions to complete the task to the best of your abilities.
Here are your skillsets:
[MASTERSTORY]:NarrStrct(StryPlnng,Strbd,ScnSttng,Exps,Dlg,Pc)-CharDvlp(ChrctrCrt,ChrctrArcs,Mtvtn,Bckstry,Rltnshps,Dlg*)-PltDvlp(StryArcs,PltTwsts,Sspns,Fshdwng,Climx,Rsltn)-ConfResl(Antg,Obstcls,Rsltns,Cnsqncs,Thms,Symblsm)-EmotImpct(Empt,Tn,Md,Atmsphr,Imgry,Symblsm)-Delvry(Prfrmnc,VcActng,PblcSpkng,StgPrsnc,AudncEngmnt,Imprv)
[*DialogWrt]:(1a-CharDvlp-1a.1-Backgrnd-1a.2-Personality-1a.3-GoalMotiv)>2(2a-StoryStruc-2a.1-PlotPnt-2a.2-Conflict-2a.3-Resolution)>3(3a-DialogTech-3a.1-ShowDontTell-3a.2-Subtext-3a.3-VoiceTone-3a.4-Pacing-3a.5-VisualDescrip)>4(4a-DialogEdit-4a.1-ReadAloud-4a.2-Feedback-4a.3-Revision)
Here are your critical instructions:
Ponder each word choice carefully to present as vivid and emotional journey as is possible. Choose verbs and nouns that are both emotional and full of imagery. Load the story with the 5 senses. Aim for 50% dialog, 25% narration, 15% body language and 10% thoughts. Your goal is to put the reader in the story.
NOTES:
- Upgraded Jinja template to correct issues with Qwen 3.5s - looping, repeatings, and long thinking as well as upgrades to tools too.
- Was also trained with new improved template to further enhance operation too.
- Image processing tested and intact.
- Code generation also tested and passed.
- System prompt - even a minor one - will enhance operation, especially at lower quants.
- Untrained 40B model (expanded from 27B, not uploaded) was also stable too and works great (unexpected).
LOOPING:
- This may happen with lower quants / prompts with "not enough meat on the bone" => Add more to the prompt and/or set rep pen to 1.05 to 1.1.
- Adding a system prompt - even a single sentence - can correct this issue and bypass the need to adjust rep pen.
WILDER? Smaller?
NEED something a wee bit wilder? Unhinged? A wee bit more raw?
See this version:
For the SMALLER, more compact 21B version see:
https://huggingface.co/DavidAU/Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking
BENCHMARKS:
arc-c arc/e boolq hswag obkqa piqa wino
This model: [instruct mode]
mxfp8 0.651,0.816,0.908,...
BASE UNTUNED MODEL:
Qwen3.6-27B HERETIC (by llmfan46) [instruct mode]
mxfp8 0.644,0.788,0.902,...
Qwen3.6-27B (by Qwen) [instruct mode]
mxfp8 0.647,0.803,0.910,0.773,0.450,0.806,0.742
Note: Instruct mode will have stronger benchmarks.
See this model (instruct, also one of my fine tunes - it scores 675 on "arc" - Arc Challenge hard):
https://huggingface.co/DavidAU/Qwen3.5-27B-Claude-4.6-OS-INSTRUCT
SAFETY ALIGNMENT:
It is gone. No nanny, no strings, no limits.
Information about this model from Qwen:
NOTE: The 40B model was built using Qwen 3.6 27B.
Qwen3.6-27B
[!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.
Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience.
Qwen3.6 Highlights
This release delivers substantial upgrades, particularly in
- Agentic Coding: the model now handles frontend workflows and repository-level reasoning with greater fluency and precision.
- Thinking Preservation: we've introduced a new option to retain reasoning context from historical messages, streamlining iterative development and reducing overhead.
For more details, please refer to our blog post Qwen3.6-27B.
Model Overview
- Type: Causal Language Model with Vision Encoder
- Training Stage: Pre-training & Post-training
- Language Model
- Number of Parameters: 27B
- Hidden Dimension: 5120
- Token Embedding: 248320 (Padded)
- Number of Layers: 64
- Hidden Layout: 16 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
- Gated DeltaNet:
- Number of Linear Attention Heads: 48 for V and 16 for QK
- Head Dimension: 128
- Gated Attention:
- Number of Attention Heads: 24 for Q and 4 for KV
- Head Dimension: 256
- Rotary Position Embedding Dimension: 64
- Feed Forward Network:
- Intermediate Dimension: 17408
- 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
For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API.
Serving Qwen3.6
Qwen3.6 can be served via APIs with popular inference frameworks. In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.6 models.
[!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 Qwen3.6 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>=0.5.10 is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
uv pip install 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/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3Tool Use: To support tool use, you can use the following command.
python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --tool-call-parser qwen3_coderMulti-Token Prediction (MTP): The following command is recommended for MTP:
python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --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
For detailed deployment guide, see the SGLang Qwen3.5 Cookbook.
vLLM
vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.
vllm>=0.19.0 is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
uv pip install vllm --torch-backend=auto
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.
vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3Tool Call: To support tool use, you can use the following command.
vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coderMulti-Token Prediction (MTP): The following command is recommended for MTP:
vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --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/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only
For detailed deployment guide, see the vLLM Qwen3.5 Recipe.
KTransformers
KTransformers is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing. For running Qwen3.6 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 Qwen3.6:
pip install "transformers[serving]"
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 Qwen/Qwen3.6-27B --port 8000 --continuous-batching
Using Qwen3.6 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"
[!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=0.0, 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:
temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0Please note that the support for sampling parameters varies according to inference frameworks.
[!Important] Qwen3.6 models operate in thinking mode by default, generating thinking content signified by
<think>\n...</think>\n\nbefore producing the final responses. To disable thinking content and obtain direct response, refer to the examples here.
Text-Only Input
from openai import OpenAI
# Configured by environment variables
client = OpenAI()
messages = [
{"role": "user", "content": "Type \"I love Qwen3.6\" backwards"},
]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=81920,
temperature=1.0,
top_p=0.95,
presence_penalty=0.0,
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) $1+\\sqrt{2}$"
}
]
}
]
response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=81920,
temperature=1.0,
top_p=0.95,
presence_penalty=0.0,
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": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?"
}
]
}
]
# 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.
response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=81920,
temperature=1.0,
top_p=0.95,
presence_penalty=0.0,
extra_body={
"top_k": 20,
"mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},
},
)
print("Chat response:", chat_response)
Instruct (or Non-Thinking) Mode
[!Important] Qwen3.6 does not officially support the soft switch of Qwen3, i.e.,
/thinkand/nothink.
Qwen3.6 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/Qwen3.6/demo/RealWorld/RealWorld-04.png"
}
},
{
"type": "text",
"text": "Where is this?"
}
]
}
]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
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)
[!Note] If you are using APIs from Alibaba Cloud Model Studio, in addition to changing
model, please use"enable_thinking": Falseinstead of"chat_template_kwargs": {"enable_thinking": False}.
Preserve Thinking
By default, only the thinking blocks generated in handling the latest user message is retained, resulting in a pattern commonly as interleaved thinking.
Qwen3.6 has been additionally trained to preserve and leverage thinking traces from historical messages.
You can enable this behavior by setting the preserve_thinking option:
from openai import OpenAI
# Configured by environment variables
client = OpenAI()
messages = [...]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=32768,
temperature=0.6,
top_p=0.95,
presence_penalty=0.0,
extra_body={
"top_k": 20,
"chat_template_kwargs": {"preserve_thinking": True},
},
)
print("Chat response:", chat_response)
[!Note] If you are using APIs from Alibaba Cloud Model Studio, in addition to changing
model, please use"preserve_thinking": Trueinstead of"chat_template_kwargs": {"preserve_thinking": False}.
This capability is particularly beneficial for agent scenarios, where maintaining full reasoning context can enhance decision consistency and, in many cases, reduce overall token consumption by minimizing redundant reasoning. Additionally, it can improve KV cache utilization, optimizing inference efficiency in both thinking and non-thinking modes.
Agentic Usage
Qwen3.6 excels in tool calling capabilities.
Qwen-Agent
We recommend using Qwen-Agent to quickly build Agent applications with Qwen3.6.
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': 'qwen3.6-27b',
'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,
'preserve_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/Qwen3.6-27B',
# '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, 'preserve_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
Qwen3.6 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.jsonfile, change therope_parametersfields intext_configto:{ "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 useVLLM_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 1010000For
sglangandktransformers, you can useSGLANG_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_parametersconfiguration only when processing long contexts is required. It is also recommended to modify thefactoras needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to setfactoras 2.0.
Best Practices
To achieve optimal performance, we recommend the following settings:
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=0.0,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:
temperature=0.7,top_p=0.80,top_k=20,min_p=0.0,presence_penalty=1.5,repetition_penalty=1.0
- Thinking mode for general tasks:
- For supported frameworks, you can adjust the
presence_penaltyparameter 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.
- We suggest using the following sets of sampling parameters depending on the mode and task type:
**Adequ
Sign up to read complete case studies, access detailed metrics, and unlock all use cases.
Sign up to read complete case studies, access detailed metrics, and unlock all use cases.