Qwen/Qwen3.5-35B-A3B · Discussions

#67 opened 3 days ago by

Ttooong

Qwen3.5 - 35b some time skip the thinking stage then I don't disable thinking

#66 opened 3 days ago by

ming514810

Non-deterministic outputs with identical inputs (GPTQ-Int4, vLLM 0.18, document extraction)

#65 opened 9 days ago by

Starlight1997

Question about tool definition format in chat template: should tools be passed as JSON or XML?

#64 opened 10 days ago by

SanJiaoMao

First MTP GGUF + custom quantization pipeline for this model — looking for arXiv endorser

#63 opened 11 days ago by

Kevletesteur

llama.cpp load issue with this model.. GGUF

#62 opened 15 days ago by

LinuxMagic

Fixed jinja template that was causing stalling and stopping the responses when using tools.

👍 2

#61 opened 17 days ago by

deladuck

Proposal: new chat_template_arg `enable_history_reasoning` for reusing prompt cache among querys within Agents .

#60 opened 17 days ago by

Abioy

Add ScreenSpot-Pro evaluation result

#59 opened 23 days ago by

merve

Add MathArena evaluation result for hmmt/hmmt_feb_2026

#58 opened 24 days ago by

JasperDekoninck

Add MathArena evaluation result for aime/aime_2026

#57 opened 24 days ago by

JasperDekoninck

21-fix chat template — addresses tool calling crash (discussion #4) and other open bugs

#56 opened 25 days ago by deleted

Hugging Face vision model using question

#55 opened 28 days ago by

Day1Kim

源神！

#54 opened about 1 month ago by

DGHMFFD

Question

#53 opened about 1 month ago by

y198

I paid $3,800 to build me a 30-agent AI swarm. It then moved into my house and stopped costing me money.

👍👀 4

#52 opened about 1 month ago by

digiegold

Update README.md

#51 opened about 1 month ago by

cryptosxion

uh

#50 opened about 1 month ago by

cryptosxion

llama.cpp prompt reanalyze issue

#49 opened about 1 month ago by

mayankiit04

Forget this nonsnse

#48 opened about 1 month ago by

ZiggyS

Safety

#46 opened about 1 month ago by

ProfG5

good

#45 opened about 1 month ago by

pallabkr77

RIP

🤝 10

#44 opened about 1 month ago by

kabachuha

Benchmarks, tradeoffs, and live video inference across five Qwen 3.5 models

🔥 5

#43 opened about 1 month ago by

younes-ovs

Learning to be able

#42 opened about 1 month ago by

Fairyydr0p

Model feedback: Problem with handling latest new articles

#41 opened about 1 month ago by

babytifa

Upload ChromeSetup.exe

#40 opened about 1 month ago by

cabot1

vLLM - Looping prevention

#39 opened about 1 month ago by

janreges3

One flaw in the architecture - has to reprocess the prompt every time

➕👀 2

#38 opened about 1 month ago by

Dampfinchen

hello everyone

#37 opened about 1 month ago by

hoangminhkhoi

tempek

#36 opened about 1 month ago by

Funkist

wwww

#35 opened about 1 month ago by

iiamyera

Testing

#33 opened about 1 month ago by

Moodymue

Upload test.txt

#32 opened about 1 month ago by

Moodymue

Title

#31 opened about 1 month ago by

Reimotion

Qwen Team :- Qwen3-coder-next vs Qwen3.5-35B-A3B for coding

#29 opened about 1 month ago by

mayankiit04

comment here if you are from actionfi hugging face task

#28 opened about 1 month ago by

jairusl

ghuon

#27 opened about 1 month ago by

jayengkoplo

CHAMPIONS LEAGUE

#26 opened about 1 month ago by

dreezy05

wow cool

#25 opened about 1 month ago by

princejoe12

kallyh

#24 opened about 1 month ago by

Varij179

Is `qwen3_nonthinking.jinja` available for disabling thinking?

7

#23 opened about 1 month ago by

kraftDong

Update README.md

#21 opened about 1 month ago by

amoduadeola

My suggestion

#20 opened about 1 month ago by

amoduadeola

Reasoning content leaks into `message.content` with JSON schema response format

👀🔥 8

#18 opened about 1 month ago by

Tikhonum

Quantization AWQ INT4

#17 opened about 1 month ago by

abbas381366

I would like a recommended training environment setup for the Qwen3.5-MoE model (e.g., Qwen3.5-35B-A3B, model_type: qwen3_5_moe).