Qwen/QwQ-32B · Discussions

How does LiveCodeBench test?

#58 opened 1 day ago by

cizhenshi

Create test

#57 opened 1 day ago by

Amyww

Upload 99ed5d4766696bd4ebc26e5d9c23e982.png

1

#56 opened 1 day ago by

axingd

Best practice for QwQ-32B evaluation

1

#55 opened 3 days ago by

wangxingjun778

Create test.txt

#54 opened 3 days ago by

xxxx443117

How to continue pt / sft on this model，any suggestions?

3

#53 opened 3 days ago by

Ken0102030405

Infinite repetitive thinking for this case:

1

#52 opened 4 days ago by

zhaocc1106

Is QwQ-32B used for custom NER extraction? Does this model work better compared to other open-source BERT transformers?

1

#51 opened 4 days ago by

amiirhmza

Budget forcing?

#50 opened 5 days ago by

mwettach

Allow prefilling assistant message

1

#49 opened 6 days ago by

tomasmcm

Update README.md

#48 opened 6 days ago by

Bschleter

Day of the week

#47 opened 6 days ago by

jac-jim

Intermittent CUDA error with model.generate() using device_map="auto" and 3 GPUs

#46 opened 7 days ago by

lucmaz98

Create Call Center Tunaiku 0818836245

#45 opened 7 days ago by

Jokiio

Does Macbook M1 max 64GB run this model well?

1

#44 opened 7 days ago by

mrk83

Too many "cross-validate" and "another method"

1

#43 opened 7 days ago by

AaronFeng753

RuntimeError: Error(s) in loading state_dict for Qwen2ForCausalLM:

#42 opened 7 days ago by

XuehangCang

8GB GPU can run this,10t/s

2

#41 opened 7 days ago by

wqerrewetw

When answering questions in Chinese, the model frequently terminates prematurely (outputs the end token). Is this a common problem?

1

#40 opened 7 days ago by

zhangw355

Refining QWQ Model Output: Direct Responses Without Step-by-Step Reasoning

1

#39 opened 7 days ago by

gslinx

It's challenging for QwQ to generate long codes...

2

#38 opened 8 days ago by

DXBTR74

Nice work... Cant-believe-its-just-32B-performance even with various different tones system prompt.

#37 opened 8 days ago by

imoc

function call时有办法跳过think吗？

1

#36 opened 8 days ago by

zhaocc1106

Failed to parse Jinja template:

2

#35 opened 8 days ago by

Vicnent

Obligatory question about model sizes...

#34 opened 8 days ago by

MrDevolver

This model beats Qwen Max!

4

#33 opened 8 days ago by

MrDevolver

remove part about long context modifications

#32 opened 8 days ago by

nbroad

add a reasoning effort option

1

#31 opened 8 days ago by

TheBigBlockPC

用vllm时应该是什么参数

6

#30 opened 8 days ago by

daiwk

遇到复杂问题时，开始推理时有<think>，推理结束了还没有</think>

6

#29 opened 8 days ago by

digits12

Is this model native 128K context length, or YaRN extended?

4

#28 opened 8 days ago by

danielhanchen

docs: update README.md

#27 opened 8 days ago by

eltociear

Thanks a lot for sharing this model!

#26 opened 8 days ago by

FalconNet

Doesn't Generate `<think>` tags

3

#25 opened 8 days ago by

bingw5

【乱码问题】输入约1w长度，输出超过1000长度，结尾部分会乱码。稳定复现！

1

#24 opened 8 days ago by

chizhu

是否需要添加系统prompt

1

#23 opened 8 days ago by

wphtrying

A pure C++ high-performance OpenAI LLM service powered by TensorRT-LLM and GRPS, with support for QWQ.

#22 opened 8 days ago by

zhaocc1106

复杂推理进入死循环

30

#21 opened 8 days ago by

frankgxy

I've tried it, the locally deployed 8G4060 can run, it's just a bit slower.

2

#20 opened 8 days ago by

Alta007

Off the charts - GREAT JOB!

1

#17 opened 9 days ago by

DavidAU

Where can I see the GPQA for this model?

#16 opened 9 days ago by

capgrey

issue of "think too much" ，how to？(chinese)

2

#14 opened 9 days ago by

fenglui

Local Installation Video - Step-by-Step Tutorial

#13 opened 9 days ago by

fahdmirzac

Tool-Calling Format

1

#12 opened 9 days ago by

G-reen

我的3090TI 24GB显存运行非常愉快！感谢开发团队！

6

#11 opened 9 days ago by

ubergarm

add missing special tokens in tokenizer.json

#10 opened 9 days ago by

mmoskal

Is `rms_norm_eps` 1e-5 or 1e-6

#9 opened 9 days ago by

danielhanchen

Oh My NVDA Position

1

#8 opened 9 days ago by

Eric2i

What languages were you trained in?

2

#7 opened 9 days ago by

NickyNicky

You guys are the pioneers!

#6 opened 9 days ago by

owao

How does LiveCodeBench test?

Create test

Upload 99ed5d4766696bd4ebc26e5d9c23e982.png

Best practice for QwQ-32B evaluation

Create test.txt

How to continue pt / sft on this model，any suggestions?

Infinite repetitive thinking for this case:

Is QwQ-32B used for custom NER extraction? Does this model work better compared to other open-source BERT transformers?

Budget forcing?

Allow prefilling assistant message

Update README.md

Day of the week

Intermittent CUDA error with model.generate() using device_map="auto" and 3 GPUs

Create Call Center Tunaiku 0818836245

Does Macbook M1 max 64GB run this model well?

Too many "cross-validate" and "another method"

RuntimeError: Error(s) in loading state_dict for Qwen2ForCausalLM:

8GB GPU can run this,10t/s

When answering questions in Chinese, the model frequently terminates prematurely (outputs the end token). Is this a common problem?

Refining QWQ Model Output: Direct Responses Without Step-by-Step Reasoning

It's challenging for QwQ to generate long codes...

Nice work... Cant-believe-its-just-32B-performance even with various different tones system prompt.

function call时有办法跳过think吗？

Failed to parse Jinja template:

Obligatory question about model sizes...

This model beats Qwen Max!

remove part about long context modifications

add a reasoning effort option

用vllm时应该是什么参数

遇到复杂问题时，开始推理时有<think>，推理结束了还没有</think>

Is this model native 128K context length, or YaRN extended?

docs: update README.md

Thanks a lot for sharing this model!

Doesn't Generate `<think>` tags

【乱码问题】输入约1w长度，输出超过1000长度，结尾部分会乱码。稳定复现！

是否需要添加系统prompt

A pure C++ high-performance OpenAI LLM service powered by TensorRT-LLM and GRPS, with support for QWQ.

复杂推理进入死循环

I've tried it, the locally deployed 8G4060 can run, it's just a bit slower.

Off the charts - GREAT JOB!

Where can I see the GPQA for this model?

issue of "think too much" ，how to？(chinese)

Local Installation Video - Step-by-Step Tutorial

Tool-Calling Format

我的3090TI 24GB显存运行非常愉快！感谢开发团队！

add missing special tokens in tokenizer.json

Is `rms_norm_eps` 1e-5 or 1e-6

Oh My NVDA Position

What languages ​​were you trained in?

You guys are the pioneers!

What languages were you trained in?