All issues

Search results

eleuther_eval recipe: unpinned GPT-2 fetch-and-discard; multimodal path silently scores empty answers; generation budget hardcoded to 256 ignoring max_gen_toks

Status: Open.

#2990 In meta-pytorch/torchtune;

· AUTHENSOR opened on Sep 3, 2026

[Bug] Torchtune recipes crash on Apple Silicon (MPS) due to hardcoded CUDA/CPU memory and device assumptions

Status: Open.

#2989 In meta-pytorch/torchtune;

· AdharshRajaram opened on Sep 3, 2026

Add support to IBM Granite models

Status: Open.

#2984 In meta-pytorch/torchtune;

· jesuino opened on Aug 12, 2026

Arbitrary code execution via the _component_ field in an untrusted recipe config

Status: Open.

#2971 In meta-pytorch/torchtune;

· geo-chen opened on Jul 13, 2026

Backport workflow-hardening fix (unpinned-uses) to 5 release branches

Status: Open.

#2968 In meta-pytorch/torchtune;

· CharlieMCY opened on Jun 29, 2026

[Proposal] Native support for SCAO (2nd-Order Optimizer) for memory-bound QLoRA recipes

Status: Open.

#2962 In meta-pytorch/torchtune;

· whispering3 opened on Apr 24, 2026

Feature : (Diagnostic) Automated FSDP Compute–Communication Overlap Diagnostic in Profiler

Status: Open.

#2954 In meta-pytorch/torchtune;

· pbpatre opened on Feb 20, 2026

kagglehub changed function: set_kaggle_credentials renamed to set_kaggle_api_token

bugSomething isn't workingSomething isn't working

Status: Open.

#2946 In meta-pytorch/torchtune;

· gabrielhuang opened on Jan 23, 2026

Bug in async_grpo_full_finetune recipe: TrainingWorker receives identical advantages

bugSomething isn't workingSomething isn't working

Status: Open.

#2943 In meta-pytorch/torchtune;

· jiatong-yu opened on Jan 4, 2026

When will the qwen3-vl model be supported?

enhancementNew feature or requestNew feature or request

Status: Open.

#2939 In meta-pytorch/torchtune;

· Cooper-Ji opened on Nov 25, 2025

After performing quantization on llama3_2 using tune run quantize --config custom_quantization.yaml, how should I proceed with inference?

Status: Open.

#2935 In meta-pytorch/torchtune;

· Begoogh opened on Oct 29, 2025

Why torchtune QAT don't support more than 1 node?

enhancementNew feature or requestNew feature or request

Status: Open.

#2932 In meta-pytorch/torchtune;

· felixslu opened on Oct 13, 2025