2.24.0 (2026-09-16) Features Add credential APIs resource to gaos. (d301255) Add RetrievalCallStep and RetrievalResultStep to interactions schema and SDKs. (9052efd) Implement environments files upload and download across Python and TypeScript SDKs (fc0834d) Wire credential APIs resource to sdk. (bf0d457) Bug Fixes Do not run functions once the automatic function calling budget is spent (2580638)…
Release 0.155.0-alpha.13
What's Changed llama.cpp updates Full Changelog: v0.34.1...v0.34.2-rc0
Verify Docker Image Signature All LiteLLM Docker images are signed with cosign. Every release is signed with the same key introduced in commit 0112e53. Verify using the pinned commit hash (recommended): A commit hash is cryptographically immutable, so this is the strongest way to ensure you are using the original signing key: cosign verify \ --key https://raw.githubusercontent.com/BerriAI/litellm/…
1.2.12 (2026-09-15) Full Changelog: v1.2.11...v1.2.12 Bug Fixes training-sdk: stop sending the obsolete managed-rollout deployment annotation training-sdk: wait for control-plane readiness when reattaching rollout deployments training-sdk: avoid copying dense tensor values when sizing requests for every client training-sdk: automatically negotiate optimized training requests for dedicated and serv…
Previously, when a member used all the Copilot AI credits available to them, they were blocked from Copilot features that consume credits. This release adds a flow for them to… The post Copilot budget increase requests are generally available appeared first on The GitHub Blog.
Release 0.155.0-alpha.2.6
TP: fix split state and granularity for fused QKV gemma4, qwen35 (#28965) model: calculate split states for attn_qkv from n_head * n_embd_head_k required for gemma4 with --fuse-qkv, where n_embd is 5376 but Q is 8192. model: handle fused full attention layers for qwen35/qwen35moe model: add TODO: [TAG_SPLIT_QGATE_QWEN] Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/…
Release 0.155.0-alpha.12
Enable CUDA graph for MTP draft (#28549) Improve CUDA graph usage for MTP Rename field Address review feedback Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/47982528 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ubuntu x64 (CPU) Ubuntu arm64 (CPU) Ubuntu s390x (CPU…
Changelog New Features: Azure OpenAI Responses: Added AzureOpenAIResponses to use the Responses API with Azure OpenAI deployments. See cookbook. Elasticsearch: Added Elasticsearch vector database with vector, keyword and hybrid search. See cookbook. DocumentationMarkdown: Added a transform for Knowledge.sync_pages that converts Mintlify and Fumadocs components into plain Markdown. See cookbook. Im…
hexagon: Support for K-Quants Q4_K and Q6_K (#28994) implement q6k/q4k kernels Squashed from: feat: implement q6k kernel hex-q6k: improve unpack accuracy hex-q4_k: add support for Q4_K kernels Co-authored-by: Max Krasnyansky maxk@qti.qualcomm.com Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/47977588 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple S…
Queued messages can now be sent into the running turn with Enter. While the agent is working, pressing Enter with an empty composer steers the first message in the queue, so it is picked up at the next turn boundary instead of waiting for the whole task to finish; the composer placeholder says so whenever the queue is non-empty. The queue head is claimed atomically, so a fast second Enter can't st…
Release 0.155.0-alpha.11
[inductor][cpu] Opt-in bmm decomposition for an unprovable batch dim …
Changes since langchain==1.4.0 release(langchain): 1.4.1 (#40498) fix(langchain): preserve open MCP object arguments (#40414) fix(langchain): correct InterruptOnConfig documentation (#40140)
OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.
hexagon: accept the zeroed rope probe in supports_op (#28995) llama probes weight placement with a rope where all params are 0, so rejecting n_dims == 0 or freq_base == 0 puts rope_freqs on the CPU. That splits the decode graph at every full-attention layer (gemma-4-E2B: 5 splits instead of 2). Assisted-by: Claude Opus 5 Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cp…
OpenAI and AARP are bringing free, hands-on ChatGPT workshops to 1,000 older adults across 10 U.S. cities to build practical AI skills safely.
DTensor's sharding propagator has no strategy for aten.miopen_batch_norm, so F.batch_norm on a sharded 4D input fails on ROCm with "Operator aten.miopen_batch_norm.default does not have a sharding strategy registered". CUDA never hits the equivalent because aten.cudnn_batch_norm has a decomposition in torch/_decomp, which DTensor's DecompShardingStrategy falls back to. The ROCm decomposition exist…
model : add support for HrmTextForCausalLM (DFM Mimir 1B) (#27625) model : add support for HrmTextForCausalLM (DFM Mimir 1B) HRM-Text runs two transformer stacks (low, high) in an alternating cycle over the same token stream. The low-cycle state z_l starts from a learned [n_embd] tensor and is broadcast over positions. conversion: new writer for the fused gqkv projection (order gate,q,k,v) remappe…
[ROCm][inductor] gfx1250 TDM support for generic descriptor codegen (…
🐤 Canary Build — v2.2.18-canary.18 Automated canary build from canary branch. Commit Information Based on changes since v2.2.18-canary.17 Commit count: 2 101c110474 🐛 fix(desktop): stop crashing the main process when a user folder cannot be resolved (#19598) (Innei) 5c12148b8e 🐛 fix(auth): allow Android passkey origins (#19640) (Tsuki) ⚠️ Important Notes This is an automated canary build and is NO…
CUDA/HIP: improve access patterns in im2col (#28013) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/47937267 macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ubuntu x64 (CPU) Ubuntu arm64 (CPU) Ubuntu s390x (CPU) Ubuntu x64 (Vulkan) Ubuntu arm64 (Vulkan) Ubuntu x64 (CU…
spacemit : fix wrong transpose function for int16 data (#25161) The sizeof(int16_t) branch in permute_transpose_impl calls rvv_transposed_s32_mn_to_nm instead of rvv_transposed_s16_mn_to_nm. This is a copy-paste bug from the sizeof(int32_t) branch above it. The s32 function uses 32-bit segment load/stores (vssseg8e32.v) on 16-bit data, reading 2x bytes per element and producing completely wrong tr…
Studio: keep the GPU order the user asked for instead of re-emitting … …it ascending (#11034) * Studio: keep the GPU order the user asked for instead of re-emitting it ascending A multi-GPU CUDA launch built the child's CUDA_VISIBLE_DEVICES from gpu_indices, which every producer sorts, so a parent CUDA_VISIBLE_DEVICES=1,0 reached llama-server as 0,1. A numeric mask carries enumeration ORDER as wel…
v1.23.3 - 2026.09.17 Downloads macOS (Apple Silicon) macOS (Intel) Windows (x64) Linux (x64 DEB) Linux (ARM64 DEB) Android What’s Changed When a reply is cut off, you'll see why and how to continue Fixed thinking controls for DeepSeek V4.1 Flash Added Turkish as an interface language Image generation model pickers now show model icons Improved OpenCode Go and Zen connectivity on mobile and the web…
Release 0.155.0-alpha.2.5
rpc : invalidate cached compute graph when a referenced buffer is freed (#24292) The server caches the most recent compute graph per device so that GRAPH_RECOMPUTE can re-execute it without resending tensor data. The cached graph nodes hold direct pointers to backend buffers that were live at graph_compute() time. If any of those buffers is later released via FREE_BUFFER, the next GRAPH_RECOMPUTE…
Per-provider model catalog files. Updated weekly by CI.