46 Releases published by 1 person

• b5957

  published Jul 22, 2025

• b5958

  published Jul 22, 2025

• b5959

  published Jul 22, 2025

• b5960

  published Jul 22, 2025

• b5961

  published Jul 22, 2025

• b5962

  published Jul 22, 2025

• b5963

  published Jul 22, 2025

• b5965

  published Jul 23, 2025

• b5966

  published Jul 23, 2025

• b5967

  published Jul 23, 2025

• b5968

  published Jul 23, 2025

• b5970

  published Jul 23, 2025

• b5972

  published Jul 23, 2025

• b5973

  published Jul 23, 2025

• b5975

  published Jul 24, 2025

• b5976

  published Jul 24, 2025

• b5978

  published Jul 24, 2025

• b5979

  published Jul 24, 2025

• b5980

  published Jul 24, 2025

• b5981

  published Jul 24, 2025

• b5984

  published Jul 24, 2025

• b5985

  published Jul 24, 2025

• b5986

  published Jul 25, 2025

• b5987

  published Jul 25, 2025

• b5988

  published Jul 25, 2025

• b5989

  published Jul 25, 2025

• b5990

  published Jul 25, 2025

• b5992

  published Jul 25, 2025

• b5993

  published Jul 25, 2025

• b5994

  published Jul 25, 2025

• b5995

  published Jul 26, 2025

• b5996

  published Jul 26, 2025

• b5997

  published Jul 26, 2025

• b5998

  published Jul 27, 2025

• b5999

  published Jul 27, 2025

• b6000

  published Jul 27, 2025

• b6001

  published Jul 27, 2025

• b6002

  published Jul 27, 2025

• b6011

  published Jul 28, 2025

• b6012

  published Jul 28, 2025

• b6013

  published Jul 28, 2025

• b6014

  published Jul 28, 2025

• b6015

  published Jul 28, 2025

• b6016

  published Jul 28, 2025

• b6017

  published Jul 28, 2025

• b6018

  published Jul 28, 2025

59 Pull requests merged by 39 people

• opencl: add ops docs

  #14910 merged Jul 28, 2025

• Extend test case filtering

  #14865 merged Jul 28, 2025

• llama-bench : use local GPUs along with RPC servers

  #14917 merged Jul 28, 2025

• ggml-cpu : deduplicate scalar implementations

  #14897 merged Jul 28, 2025

• SYCL: Add set_rows support for quantized types

  #14883 merged Jul 28, 2025

• mtmd : add support for Voxtral

  #14862 merged Jul 28, 2025

• CUDA: fix pointer incrementation in FA

  #14916 merged Jul 28, 2025

• Add support for SmallThinker model series

  #14898 merged Jul 28, 2025

• sycl: refactor quantization to q8_1

  #14815 merged Jul 28, 2025

• ops : update BLAS

  #14914 merged Jul 28, 2025

• ops : update Metal

  #14912 merged Jul 28, 2025

• sync : ggml

  #14911 merged Jul 28, 2025

• quantize: update README.md

  #14905 merged Jul 27, 2025

• Vulkan: add ops docs

  #14900 merged Jul 27, 2025

• SYCL: add ops doc

  #14901 merged Jul 27, 2025

• llama : clarify comment about pp and tg graphs [no ci]

  #14895 merged Jul 27, 2025

• vulkan : add fp16 support for the conv_2d kernel

  #14872 merged Jul 27, 2025

• vulkan: skip empty set_rows to avoid invalid API usage

  #14860 merged Jul 27, 2025

• make rope_yarn_log_mul optional for deepseek2

  #14896 merged Jul 27, 2025

• Fix kq_scale for the attention layers of PLaMo2

  #14892 merged Jul 27, 2025

• Docs: add instructions in ops.md + simplify backend csv

  #14889 merged Jul 27, 2025

• HIP: Enable Matrix cores for MMQ Kernels, Enable stream-K for CDNA 3

  #14624 merged Jul 26, 2025

• CANN: Implement GLU ops

  #14884 merged Jul 26, 2025

• musa: fix build warnings (unused variable)

  #14869 merged Jul 26, 2025

• ggml-cpu: disable GGML_NNPA by default due to instability

  #14880 merged Jul 25, 2025

• metal: SSM_SCAN performance

  #14743 merged Jul 25, 2025

• opencl: add fused rms_norm_mul

  #14841 merged Jul 25, 2025

• docs: update HOWTO‑add‑model.md for ModelBase and new model classes

  #14874 merged Jul 25, 2025

• Code health: Remove invalid portPos specifiers from graph dumping to dot files

  #14838 merged Jul 25, 2025

• context : restore preemptive sched reset when LLAMA_SET_ROWS=0

  #14870 merged Jul 25, 2025

• mtmd : Fix 32-bit narrowing issue in export-lora and mtmd clip

  #14503 merged Jul 25, 2025

• GGML: Check for null buffers in get/set/copy tensor RPC endpoints

  #14868 merged Jul 25, 2025

• sched : fix multiple evaluations of the same graph with pipeline parallelism

  #14855 merged Jul 25, 2025

• musa: upgrade musa sdk to rc4.2.0

  #14498 merged Jul 24, 2025

• sync : ggml

  #14858 merged Jul 24, 2025

• context : perform output reorder lazily upon access after sync

  #14853 merged Jul 24, 2025

• chat : fix kimi-k2 chat template

  #14852 merged Jul 24, 2025

• sycl: unified semantics of block offset calculation

  #14814 merged Jul 24, 2025

• fix: restore MiniCPM inference after Granite Four changes

  #14850 merged Jul 24, 2025

• docs: add libcurl-dev install hint for Linux distros

  #14801 merged Jul 24, 2025

• metal : fix fusion across different encoders

  #14849 merged Jul 24, 2025

• sycl: fix undefined variable in work group size check

  #14843 merged Jul 24, 2025

• convert: text-only support for GLM-4.1V-9B-Thinking

  #14823 merged Jul 23, 2025

• CUDA: fix overflow in FA, tune performance

  #14840 merged Jul 23, 2025

• CUDA: fix compilation with GGML_CUDA_F16

  #14837 merged Jul 23, 2025

• ci : correct label refactor->refactoring

  #14832 merged Jul 23, 2025

• tests : add non-cont K,V FA tests

  #14756 merged Jul 23, 2025

• CUDA: fix quantized KV cache + multiple sequences

  #14822 merged Jul 23, 2025

• bug fix: handle saving/loading null layers in recurrent memory

  #14675 merged Jul 23, 2025

• ggml: fix loongarch quantize_row_q8_1 error

  #14827 merged Jul 23, 2025

• [CANN] weight format to nz for Ascend310P3

  #14407 merged Jul 23, 2025

• CUDA: add fused rms norm

  #14800 merged Jul 23, 2025

• model card yaml tab->2xspace

  #14819 merged Jul 22, 2025

• vulkan: fix rms_norm_mul to handle broadcasting dim0

  #14817 merged Jul 22, 2025

• llama : add model type detection for rwkv7 7B&14B

  #14816 merged Jul 22, 2025

• imatrix: add option to display importance score statistics for a given imatrix file

  #12718 merged Jul 22, 2025

• Mtmd: add a way to select device for vision encoder

  #14236 merged Jul 22, 2025

• cuda : implement bf16 cpy ops and enable bf16 cont

  #14763 merged Jul 22, 2025

• opencl: remove unreachable return

  #14806 merged Jul 22, 2025

26 Pull requests opened by 19 people

• opencl: tiled mul_mat with local memory for f16 and f32

  #14809 opened Jul 22, 2025

• convert : handle pre-quantized models

  #14810 opened Jul 22, 2025

• feat(batched): Add functionality to upload benchmark test results

  #14811 opened Jul 22, 2025

• graph : reduce splits for recurrent and hybrid models

  #14825 opened Jul 23, 2025

• test-backend-ops: enables perf/eval testing of composite ops

  #14833 opened Jul 23, 2025

• SvelteKit-based WebUI

  #14839 opened Jul 23, 2025

• imatrix : use GGUF by default

  #14842 opened Jul 24, 2025

• Adding chat template support for Granite model

  #14864 opened Jul 24, 2025

• Support intern-s1

  #14875 opened Jul 25, 2025

• model: add hunyuan dense

  #14878 opened Jul 25, 2025

• GGML: Fix leak of backend buffer memory address in RPC

  #14882 opened Jul 26, 2025

• imatrix: calculate activation-based statistics for new format (GGUF) imatrices

  #14891 opened Jul 26, 2025

• Vulkan: Fix minor debug mode issues

  #14899 opened Jul 27, 2025

• Vulkan: Add Integer Dot Product mul_mat_vec shader for legacy quants

  #14903 opened Jul 27, 2025

• ggml : repack block_iq4_nlx8 (AVX)

  #14904 opened Jul 27, 2025

• cuda : add softcap fusion

  #14907 opened Jul 27, 2025

• opencl: fixed a typo

  #14908 opened Jul 27, 2025

• repack : optimize mul_mat_id path

  #14918 opened Jul 28, 2025

• CUDA: add roll

  #14919 opened Jul 28, 2025

• CUDA: skip masked KV slices for all FA kernels

  #14924 opened Jul 28, 2025

• bench, common : add CPU extra buffer types

  #14925 opened Jul 28, 2025

• embeddings: fix extraction of CLS pooling results

  #14927 opened Jul 28, 2025

• mtmd : support home-cooked Mistral Small Omni

  #14928 opened Jul 28, 2025

• server-bench: make seed choice configurable

  #14929 opened Jul 28, 2025

• HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path.

  #14930 opened Jul 28, 2025

• HIP: Ignore unsupported unroll transformation in fattn-vec

  #14931 opened Jul 28, 2025

41 Issues closed by 19 people

• Compile bug: Looking for C++ include rocwmma/rocwmma.hpp - not found

  #14538 closed Jul 28, 2025

• Eval bug: Unable to run with Qwen3 model on rocm with gfx1100, but works on cpu

  #14696 closed Jul 28, 2025

• Misc. bug: slow model loading to GPU when size > 64GB (Vulkan)

  #14854 closed Jul 28, 2025

• Eval bug: Repeated sequences with gemma3 and image recognition

  #14888 closed Jul 28, 2025

• Feature Request: Support EXAONE 4.0

  #14474 closed Jul 28, 2025

• Misc. bug: [SYCL] llama-cli built by Visual Studio 2022 is not working

  #14086 closed Jul 28, 2025

• Research: mmap eviction

  #14154 closed Jul 28, 2025

• prismatic-vlms to gguf?

  #14159 closed Jul 28, 2025

• Eval bug: MultiGPU x MultiModels = 100% GPU

  #14890 closed Jul 27, 2025

• Misc. bug: using --jinja flag with server and Qwen3 models removes thinking block, still works on llama-cli

  #14894 closed Jul 27, 2025

• ggml_vulkan: RADV crash on ggml_set_rows due to zero size buffer

  #14845 closed Jul 27, 2025

• Eval bug: gemma-3n-E4B-it-Q8_0.gguf is speaking nonsense

  #14885 closed Jul 27, 2025

• Misc. bug: llama-server webui with --jinja flag does not show thinking when using reasoning models

  #14007 closed Jul 27, 2025

• Vulkan Runner Frequent Crashing under workload

  #14105 closed Jul 27, 2025

• Eval bug: (MAC) fail in `GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_Q8_0_H96, flash_attn_ext_q8_0_h96, has_simdgroup_mm);`

  #14110 closed Jul 27, 2025

• Misc. bug: --cache-reuse no longer seems to be caching prompt prefixes

  #14113 closed Jul 27, 2025

• Misc. bug: "llama_context_params::swa_full = true" causes very large RAM/VRAM usage

  #14123 closed Jul 27, 2025

• Misc. bug: llama-server drops multi-part content for final assistant message

  #14137 closed Jul 27, 2025

• Metrics should not include : in Prometheus metric names

  #14150 closed Jul 27, 2025

• Feature Request: (webui) read data from /props endpoint and use it on the webui

  #11717 closed Jul 26, 2025

• Support Hybrid Models

  #12331 closed Jul 26, 2025

• Eval bug: s390x GGML_NNPA=ON Generates Gibberish Tokens at Different Thread Counts

  #14877 closed Jul 25, 2025

• Eval bug: Generation speed loss after b5920

  #14876 closed Jul 25, 2025

• Performance regression with multiple GPUs in commit 01612b7

  #14863 closed Jul 25, 2025

• Eval bug: LLAMA_SET_ROWS=1 gibberish output with Dual GPU offload

  #14795 closed Jul 25, 2025

• Eval bug: Unusual high RAM usage on Windows when running DeepSeek V3 Q2_K_XL/IQ2_XXS, on Hybrid CPU+GPU

  #13978 closed Jul 25, 2025

• Eval bug: MiniCPM4 0.5B run failed

  #14094 closed Jul 25, 2025

• Eval bug: Gemma3 decode and update_slots fail with parallel slots

  #14097 closed Jul 25, 2025

• Eval bug: gemma3 generates infinite "and" output after commit bf9087f

  #14835 closed Jul 24, 2025

• Misc. bug: llama-server --batch-size always set to 64

  #14046 closed Jul 24, 2025

• Eval bug: Ollama Runner uses only 1 CPU for its threads, in guaranteed mode in pod when 8 CPUs are allocated to it

  #14089 closed Jul 24, 2025

• Misc. bug: Server tests /health race conditions

  #14092 closed Jul 24, 2025

• Compile bug: convert.cu

  #14834 closed Jul 23, 2025

• Misc. bug: CUDA docker image - libcurl: file too short

  #14813 closed Jul 23, 2025

• Misc. bug: Failed to run `llama-server` when trying to recurrence the issue #14812

  #14829 closed Jul 23, 2025

• [How to serve lookahead decoding Qwen 3]

  #14057 closed Jul 23, 2025

• Eval bug: Model produces gibberish or repeated output when using `-sm row` on CUDA

  #14075 closed Jul 23, 2025

• Quantize bug: Ernie4.5 MoE 300B low-bit quantization crashes

  #14788 closed Jul 22, 2025

• Feature Request: Built-in Token Probability Output for Inference API

  #14611 closed Jul 22, 2025

• Feature Request: Direct FP8 conversion from convert_hf_to_gguf.py

  #14762 closed Jul 22, 2025

• Eval bug: RWKV inference with llama-parallel gets wrong output with lmhead offloaded to GPU

  #14211 closed Jul 22, 2025

18 Issues opened by 18 people

• Compile bug: "Illegal instruction" on StarFive VisionFive2 (risc-v)

  #14926 opened Jul 28, 2025

• Eval bug: Uncaught exception during inference crashes llama-server

  #14923 opened Jul 28, 2025

• Feature Request: GLM 4.5 MoE support

  #14921 opened Jul 28, 2025

• Refactor: Merge build_moe_ffn_from_probs function into build_moe_ffn

  #14920 opened Jul 28, 2025

• Eval bug: Unable to run Qwen3-Coder-480B-A35B-Instruct-UD-Q2_K_XL

  #14915 opened Jul 28, 2025

• Feature Request: Implement missing ops from backends

  #14909 opened Jul 28, 2025

• Eval bug: No kernel named rms_norm_f32_sycl

  #14887 opened Jul 26, 2025

• Speed regression with -fa and -ctk

  #14881 opened Jul 25, 2025

• Misc. bug:convert_hf_to_gguf.py lora fine-tuning effects for visual recognition are completely lost – only the text-related fine-tuning remains effective.

  #14867 opened Jul 25, 2025

• Eval bug: cline plugin for VS Code does not work with any GGUF

  #14866 opened Jul 25, 2025

• Eval bug: Embedding output differs significantly between b4712 and b4713

  #14848 opened Jul 24, 2025

• Misc. bug: Regression in unified KV cache appears after `llama.cpp` release b5912 in b5913

  #14847 opened Jul 24, 2025

• Please implement phi-3-M3-coder

  #14846 opened Jul 24, 2025

• Eval bug: failed to allocate compute pp buffers

  #14836 opened Jul 23, 2025

• Misc. bug: llama-server issue on Windows when compiling from source code

  #14826 opened Jul 23, 2025

• Eval bug: unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q2_K_XL using HIP backend (AMD MI300X) outputs `GGGGG`

  #14824 opened Jul 23, 2025

• Misc. bug: llama-server embedding endpoint returns vectors with just null values after a while

  #14812 opened Jul 22, 2025

• Misc. bug: Server cpp no image_data being used

  #14807 opened Jul 22, 2025

64 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.

• Add LLaDA 8b Diffusion model

  #14771 commented on Jul 28, 2025 • 14 new comments

• Improve Mistral models integration with llama.cpp

  #14737 commented on Jul 28, 2025 • 11 new comments

• Fix MinicpmV model converter and clip to avoid using hardcode.

  #14750 commented on Jul 28, 2025 • 9 new comments

• feat: Add optional prompt processing progress streaming

  #14731 commented on Jul 28, 2025 • 4 new comments

• llama-server : implement universal assisted decoding

  #12635 commented on Jul 28, 2025 • 4 new comments

• examples : predicted output for text generation

  #14739 commented on Jul 24, 2025 • 3 new comments

• Request for Official Support of AMD Ryzen AI Platform NPU

  #14377 commented on Jul 29, 2025 • 0 new comments

• Misc. bug: Saving and restoring an empty slot does a crasherino

  #14434 commented on Jul 29, 2025 • 0 new comments

• Misc. bug: OpenAI API v1/responses llama-server

  #14702 commented on Jul 28, 2025 • 0 new comments

• Misc. bug: LLAMA-SERVER is 40% slower than LLAMA-CLI when using identical parameters including -ot option for tensor offloading

  #14201 commented on Jul 28, 2025 • 0 new comments

• Feature request: Graphical GGUF viewer

  #6715 commented on Jul 28, 2025 • 0 new comments

• Compile bug: gcc-12: error: unrecognized command-line option ‘-compress-mode=size’

  #14260 commented on Jul 28, 2025 • 0 new comments

• Eval bug: Unexpected empty grammar stack after accepting piece: <unused32>

  #14413 commented on Jul 28, 2025 • 0 new comments

• bug: GGML_ASSERT(backend_embd != nullptr) failed error at llama.cpp:14775

  #14418 commented on Jul 28, 2025 • 0 new comments

• Some models like gemma-3n crashes - rocBLAS error: Cannot read /opt/rocm-6.4.1/lib/llvm/bin/../../../lib/rocblas/library/TensileLibrary.dat: No such file or directory for GPU arch : gfx1036

  #14421 commented on Jul 28, 2025 • 0 new comments

• Misc. bug:

  #14422 commented on Jul 28, 2025 • 0 new comments

• Compile bug: loop not unrolled ROCm warnings

  #14776 commented on Jul 27, 2025 • 0 new comments

• Eval bug: Gemma 3n on Vulkan on Ryzen APUs produces garbled output

  #14525 commented on Jul 27, 2025 • 0 new comments

• Feature Request: per-chat prompt caching

  #14470 commented on Jul 27, 2025 • 0 new comments

• Misc. bug: Server/Chat parallel tool calling not working

  #14101 commented on Jul 29, 2025 • 0 new comments

• Eval bug: Cannot load Qwen3 ranking models

  #13820 commented on Jul 29, 2025 • 0 new comments

• Introduce Graph Profiler

  #9659 commented on Jul 25, 2025 • 0 new comments

• Overlap CUDA graph building and processing to minimize GPU idle time and improve tokens per seconds performance.

  #11867 commented on Jul 25, 2025 • 0 new comments

• [WIP]backend: Integrating QNN (Qualcomm AI Engine Direct) as a dedicated backend for Qualcomm NPUs

  #12063 commented on Jul 22, 2025 • 0 new comments

• Introduce New Lookup-Table(LUT)-Based Matrix Multiplication Method (TMAC)

  #13206 commented on Jul 25, 2025 • 0 new comments

• CUDA: update build CTK version to 12.8

  #13360 commented on Jul 27, 2025 • 0 new comments

• finetune.cpp command-line arg

  #13873 commented on Jul 23, 2025 • 0 new comments

• convert: add eagle2 draft arch

  #13908 commented on Jul 28, 2025 • 0 new comments

• llama : support qwen3 rerank and embeddings

  #14029 commented on Jul 23, 2025 • 0 new comments

• compare-commits.sh: support both llama-bench and test-backend-ops

  #14392 commented on Jul 24, 2025 • 0 new comments

• Chore: batch prompts, extract tensors specific layer

  #14463 commented on Jul 28, 2025 • 0 new comments

• mtmd : Support jinja in libmtmd (Only for QwenVL and Qwen Omni)

  #14730 commented on Jul 22, 2025 • 0 new comments

• Feature Request: Add --upload to llama-bench

  #14791 commented on Jul 22, 2025 • 0 new comments

• Eval bug: Nondeterministic output with ROCm backend despite zero temperature

  #14727 commented on Jul 22, 2025 • 0 new comments

• Eval bug: CUDA error: operation not supported

  #14692 commented on Jul 22, 2025 • 0 new comments

• [BUG] DeepSeek V3 weight_scale_inv tensor mapping not supported in converter

  #14781 commented on Jul 22, 2025 • 0 new comments

• Eval bug: KV cache stopped working in b5554 version

  #14071 commented on Jul 23, 2025 • 0 new comments

• Misc. bug: llama-quant "found point XXX not on grid: XXXX"

  #14798 commented on Jul 23, 2025 • 0 new comments

• server : add "token healing" support

  #5765 commented on Jul 23, 2025 • 0 new comments

• Feature Request: Ability to pack multiple GGUFs into single one

  #13028 commented on Jul 23, 2025 • 0 new comments

• Refactor: (clip.cpp) identify and regroup pre-processing strategies

  #13077 commented on Jul 23, 2025 • 0 new comments

• deprecate llama_batch_get_one and llama_get_logits

  #4491 commented on Jul 23, 2025 • 0 new comments

• Eval bug: ROCml -> ggml_cuda_compute_forward: MUL_MAT failed when running unsloth/Kimi K2

  #14787 commented on Jul 23, 2025 • 0 new comments

• Feature Request: add tool calling for deepseek-r1-0528

  #14557 commented on Jul 23, 2025 • 0 new comments

• Feature Request: Suggest to provide armv7l version to run on Raspberry Pi devices.

  #14348 commented on Jul 24, 2025 • 0 new comments

• Feature Request: Support GLM-4.1V-9B-Thinking

  #14495 commented on Jul 24, 2025 • 0 new comments

• ggml_vulkan: device Vulkan0 does not support 16-bit storage.

  #14805 commented on Jul 24, 2025 • 0 new comments

• Regarding the build for 8060S (gfx1151):

  #14734 commented on Jul 24, 2025 • 0 new comments

• Misc. bug: Failure to allocate buffer with ROCm 6.4

  #14178 commented on Jul 25, 2025 • 0 new comments

• Misc. bug: RUNPATH properties are not properly set

  #13740 commented on Jul 25, 2025 • 0 new comments

• Eval bug: terminate called after throwing an instance of 'std::runtime_error' what(): Unexpected empty grammar stack after accepting piece: [control_36]

  #13690 commented on Jul 25, 2025 • 0 new comments

• Misc. bug: DeepSeek-R1 0528 671b:Q4_K_XL think tags do not close sometimes

  #14679 commented on Jul 25, 2025 • 0 new comments

• Feature Request: Add support for Kokoro TTS

  #11050 commented on Jul 25, 2025 • 0 new comments

• Feature Request: Add TPU/Hardware Accelerator Support (e.g., Google Coral, Hailo) to llama.cpp

  #11603 commented on Jul 25, 2025 • 0 new comments

• Misc. bug: missing messages in JSON export via llama-server web UI

  #13552 commented on Jul 25, 2025 • 0 new comments

• Feature Request: allow running llama with an idle (lowest) priority as well

  #14382 commented on Jul 26, 2025 • 0 new comments

• Compile error for ggml_gemv_q4_K_8x8_q8_K on Intel x86_64 MacOS (AVX2)

  #14372 commented on Jul 26, 2025 • 0 new comments

• Misc. bug: Inconsistent Gemma3 implementation in rope factor

  #14367 commented on Jul 26, 2025 • 0 new comments

• Feature Request: Add support for moonshotai/Kimi-VL-A3B-Instruct

  #14318 commented on Jul 26, 2025 • 0 new comments

• Eval bug: [CANN]AutoDL Ascend 910B instance running DeepSeek-r1 32B_Q8 error

  #14291 commented on Jul 26, 2025 • 0 new comments

• Error while converting peft finetuned merged model to gguf

  #12494 commented on Jul 26, 2025 • 0 new comments

• Eval bug: Regression: Tool calls still returned in content field as JSON string instead of tool_calls array

  #14697 commented on Jul 26, 2025 • 0 new comments

• Eval bug: Program crashes during long input inference when batch size is set to 16384

  #14325 commented on Jul 27, 2025 • 0 new comments

• mtmd: Any plan for mtmd to support video input and audio output?

  #14295 commented on Jul 27, 2025 • 0 new comments