11 Pull requests merged by 4 people

• [v0.5.0a1] add support for torch 2.8.0

  #1258 merged Aug 22, 2025

• Add v4_updated tokenizer for 300m, 1b, and 3b models

  #1257 merged Aug 22, 2025

• add support for torch 2.8.0

  #1259 merged Aug 22, 2025

• Multi split eval

  #1251 merged Aug 22, 2025

• Gilkeren/context text

  #1255 merged Aug 22, 2025

• nightly build w/ pt271+cu128

  #1254 merged Aug 21, 2025

• bump to pt271+cu128

  #1252 merged Aug 21, 2025

• Max batch size and text length filtering

  #1253 merged Aug 21, 2025

• Adding test presets for mms + support for any encoder in reference model

  #1249 merged Aug 20, 2025

• Bleu memory leak

  #1250 merged Aug 20, 2025

• Zero-shot post fixes

  #1223 merged Aug 19, 2025

3 Pull requests opened by 3 people

• mms upstreaming bugfix and docs

  #1256 opened Aug 22, 2025

• Jacklanchantin/qwen

  #1260 opened Aug 22, 2025

• startup fixes in v0.5.0a1

  #1261 opened Aug 23, 2025

4 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Main w2v2 pretraining

  #1100 commented on Aug 22, 2025 • 42 new comments

• Flex attention integration with block mask cache

  #1210 commented on Aug 19, 2025 • 0 new comments

• skywork and some qwen changes

  #1232 commented on Aug 24, 2025 • 0 new comments

• Gilkeren/debug2

  #1235 commented on Aug 18, 2025 • 0 new comments