1 Release published by 1 person

• v3.3.0

  published May 9, 2025

4 Pull requests merged by 3 people

• adjust the round_up_seq logic to align with prefill warmup phase on…

  #3224 merged May 12, 2025

• change HPU warmup logic: seq length should be with exponential growth

  #3217 merged May 10, 2025

• Prepare for 3.3.0

  #3220 merged May 9, 2025

• Chunked Prefill VLM

  #3188 merged May 6, 2025

4 Pull requests opened by 4 people

• Deepseek r1

  #3211 opened May 7, 2025

• Update to Torch 2.7.0

  #3221 opened May 10, 2025

• Refine logging for Gaudi warmup

  #3222 opened May 10, 2025

• Enable Llama4 for gaudi backend

  #3223 opened May 11, 2025

2 Issues closed by 2 people

• when HF_HUB_OFFLINE == 1, cache config becomes inconsistent

  #3212 closed May 8, 2025

• Adapt the response_format closer to OpenAIs format

  #3058 closed May 7, 2025

6 Issues opened by 5 people

• Launching a container with an unprivileged user

  #3225 opened May 12, 2025

• Docker Image that works on RTX 5090

  #3219 opened May 9, 2025

• Model request: RecurrentGemma

  #3216 opened May 9, 2025

• Phi 4 Reasoning Not able to Start

  #3215 opened May 8, 2025

• Strange output when using Structured output with Gemma 3 12b it

  #3214 opened May 8, 2025

• Whether it supports Huawei Atlas300 graphics card?

  #3213 opened May 8, 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.

• RuntimeError: Cannot load 'awq' weight when running Qwen2-VL-72B-Instruct-AWQ model

  #2944 commented on May 7, 2025 • 0 new comments

• Qwen 3 support

  #3199 commented on May 8, 2025 • 0 new comments

• Add support for phi-4-mini and phi-4-multimodal

  #3071 commented on May 9, 2025 • 0 new comments

• Function/tool calling never resolves

  #2986 commented on May 13, 2025 • 0 new comments