Search results

Unsloth + vLLM multi-GPU setup in ART crashes when restricting CUDA_VISIBLE_DEVICES or using tensor parallelism >1 on H100 GPUs

Status: Open.

#478 In OpenPipe/ART;

· ansh-info opened on Dec 11, 2025

RULER produces scores >1, arbitrary outputs (7, 10, 15…), nondeterministic behavior, and malformed JSON when judge model is Qwen3-32B running on Ollama

Status: Open.

#477 In OpenPipe/ART;

· ansh-info opened on Dec 11, 2025

RULER + Litellm with Ollama: <code>ollama_chat</code> recommended, but <code>/api/chat</code> cannot generate required JSON schema

Status: Open.

#476 In OpenPipe/ART;

· ansh-info opened on Dec 11, 2025

RULER scoring + training tightly coupled to Litellm/OpenAI, cannot cleanly use ChatOllama/ChatNVIDIA as judge/inference models

Status: Open.

#475 In OpenPipe/ART;

· ansh-info opened on Dec 11, 2025

<code>init_chat_model</code> always uses <code>ChatOpenAI</code>, ignores args, and still calls OpenAI / hits 10-minute timeout with Ollama

Status: Open.

#474 In OpenPipe/ART;

· ansh-info opened on Dec 11, 2025

Context window incorrectly capped at 8192 tokens when using Ollama (token count exceeded 8192)

Status: Open.

#473 In OpenPipe/ART;

· ansh-info opened on Dec 11, 2025

Training hangs at 0/12 with Qwen2.5-14B + Unsloth/vLLM (KV cache reported as 0.0 GB despite free VRAM)

Status: Open.

#472 In OpenPipe/ART;

· ansh-info opened on Dec 11, 2025

Is it not possible to set the rollout engine to multiple gpus?

Status: Open.

#470 In OpenPipe/ART;

· melanibe opened on Dec 10, 2025

Training hangs after a few steps

Status: Open.

#469 In OpenPipe/ART;

· melanibe opened on Dec 8, 2025

Please relax the open ai requirements <code>openai&gt;=1.65.5,&lt;=1.99.1.</code>

Status: Open.

#468 In OpenPipe/ART;

· ShravanSunder opened on Nov 29, 2025

Training gets stuck after completing just one trajectory

Status: Open.

#459 In OpenPipe/ART;

· abd-hsn opened on Nov 18, 2025

Cuda out of Memory

Status: Open.

#452 In OpenPipe/ART;

· ck-amrahd opened on Nov 9, 2025