130 wires and counting

$ follow Hugging Face

Keep up with Hugging Face in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-08-15
stories 31

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

DIFFUSERS CLOSES LORA GAPS WHILE MOON-IDE PINS CACHE REPLICAS

By RepoJournal · Filed · About Hugging Face · Composed from the cited sources · methodology

Diffusers shipped LoRA loading for MiniMax-H3 and fixed test failures across the framework, while moon-ide added session affinity to stabilize prompt cache distribution.

The diffusers team merged four major fixes this cycle. LoRA loading support for MiniMax-H3 now covers three additional published layouts that previously failed to load [1], addressing key-layout gaps and a bug where unmatched layouts silently returned without exception. In parallel, a PEFT refactor broke LoRA test warnings, which the team fixed by dropping adapter name checks from the mismatch logic [5]. SVD tests also received corrections [6]. On the ROCm front, diffusers replaced the `aiter` dependency with a Hub-hosted kernel for FlashAttention [7], reducing external dependencies.

Meanwhile, moon-ide is shipping session affinity UUIDs to pin HF prompt cache replicas and now tracks session-lifetime cache hit rates instead of per-request percentages [2] [3], improving observability for cached workloads. The team also added image attachments to the phone composer with paste and camera support [8].

TRL added vLLM 0.27.0 and 0.27.1 support with no breaking changes for TRL [4], and fixed an SFT bug where models without `past_key_values` or `attentions` in their output raised `AttributeError` on first step [9]. Across all four repos: 18 commits, 13 PRs.

Action items

References

  1. [1] Add LoRA loading support for MiniMax-H3 (#14408) ↗ huggingface/diffusers
  2. [2] Add session affinity UUID to pin HF prompt cache replicas ↗ huggingface/moon-ide
  3. [3] Track session-lifetime cache hit rate instead of per-request percentage ↗ huggingface/moon-ide
  4. [4] Add support for vLLM 0.27.0 and 0.27.1 ↗ huggingface/trl
  5. [5] FIX LoRA tests warning about unexpected keys (#14476) ↗ huggingface/diffusers
  6. [6] fix SVD tests (#14475) ↗ huggingface/diffusers
  7. [7] [FA2] Use `kernels-community/aiter-flash-attn-ck` Hub kernel for ROCm FlashAttention (drop `aiter` dependency) (#14436) ↗ huggingface/diffusers
  8. [8] Add image attachments to phone composer with paste and camera support ↗ huggingface/moon-ide
  9. [9] [SFT] Treat past_key_values and attentions as optional on the model output ↗ huggingface/trl

Quick answers

What shipped in Hugging Face on August 15, 2026?
Diffusers shipped LoRA loading for MiniMax-H3 and fixed test failures across the framework, while moon-ide added session affinity to stabilize prompt cache distribution. In total, 18 commits and 13 pull requests landed.
Who contributed to Hugging Face on August 15, 2026?
10 developers shipped this update, including coyotte508, dependabot, Quentin Gallouédec, Albert Villanova del Moral, Hakureirm, apolinário, BADAOUI Abdennacer, and Benjamin Bossan, and 2 more.
What were the notable Hugging Face updates?
Add LoRA loading support for MiniMax-H3 (#14408), Add session affinity UUID to pin HF prompt cache replicas, and Track session-lifetime cache hit rate instead of per-request percentage.