| pr |
1.00 |
docs: fix dead OpenAI compatibility link in examples |
|
"🤖 Generated with Claude Code" explicitl |
2026-07-22 |
| pr |
0.75 |
server: cache GGUF metadata to avoid re-parsing on warm requ |
|
Polite, meta-comments and AI tool mentio |
2026-07-23 |
| pr |
0.50 |
mlxrunner: continuous batching for OLLAMA_NUM_PARALLEL |
|
Some structured summary and checklist, b |
2026-07-22 |
| pr |
0.50 |
sched: tag unloadedCh with modelKey to avoid eviction crosst |
|
More structured test plan, some template |
2026-07-22 |
| pr |
0.30 |
feat(server/convert): add dynamic LoRA adapter support for n |
|
Standard changelog style; minor formalit |
2026-07-24 |
| pr |
0.30 |
feat(server): add inference webhook hooks for input/output i |
|
Slightly more formal, but still domain-f |
2026-04-16 |
| pr |
0.20 |
docs: fix broken links, convert GitHub callouts to Mintlify |
|
Standard documentation updates; changelo |
2026-03-24 |
| pr |
0.20 |
cmd: add ollama update command and upgrade guidance (#17219) |
|
Technical language with minor shorthand, |
2026-07-22 |
| pr |
0.20 |
server/upload: fix data race on blobUpload.done and err |
|
Some template structure but bug details |
2026-07-24 |
| pr |
0.20 |
model/renderers: keep qwen images separate |
|
Direct, informal, bug-oriented; detailed |
2026-07-22 |
| pr |
0.20 |
tools: support parsing tool calls with omitted arguments |
|
Technical details and bug descriptions; |
2026-07-22 |
| pr |
0.20 |
tools: stop dropping content generated after bare-JSON tool |
|
Technical bugfix explanation; jargon and |
2026-07-22 |
| pr |
0.20 |
tools: prevent findTool matching short tool names inside lon |
|
Contains code-level detail, clear bug/fi |
2026-07-22 |
| pr |
0.20 |
tools: prevent infinite loop when non-tool JSON precedes val |
|
Succinct and domain-focused; technical b |
2026-07-22 |
| pr |
0.20 |
server/sched: tag unloadedCh with modelKey to prevent evicti |
|
Highly technical, with step-by-step bug |
2026-07-22 |
| pr |
0.20 |
server/sched: guard needsReload against nil model and llama |
|
Technical concurrency bug writeup, not e |
2026-07-22 |
| pr |
0.20 |
fix: drop .local/.internal from allowedHost suffix allowlist |
|
Apologetic tone but natural and context- |
2026-07-22 |
| commit |
0.15 |
mlx: x/create rewrite (#16919) |
|
Technical breakdown with domain specific |
2026-07-04 |
| pr |
0.15 |
feat(cmd): add --json flag to list and ps commands |
|
Detailed but casual and domain-specific |
2026-07-23 |
| pr |
0.15 |
feat(cmd): add --json flag to list and ps commands |
|
Detailed but casual and domain-specific |
2026-07-23 |
| pr |
0.15 |
agent: import skills from coding agents |
|
Technical, domain-specific, concise—no A |
2026-07-21 |
| commit |
0.10 |
mlx: keep loaded model memory resident (#17367) |
|
Technical, terse, and domain-specific; n |
2026-07-24 |
| commit |
0.10 |
server: fix ps data race on scheduler loaded map (#17376) |
|
In-depth technical details, informal; si |
2026-07-24 |
| commit |
0.10 |
test: harden flaky updater and transfer unit tests (#17378) |
|
Highly specific, informal test rationale |
2026-07-24 |
| commit |
0.10 |
model/parsers: finalize incomplete GLM tool calls (#17250) |
|
Technical, detailed, but uses natural to |
2026-07-22 |
| commit |
0.10 |
model: add Laguna v8 chat support and fix Metal inference (# |
|
Technical detail, domain-specific, not A |
2026-07-21 |
| commit |
0.10 |
model: align gemma4 chat template handling (#17182) |
|
Uses domain-specific terms and natural, |
2026-07-14 |
| commit |
0.10 |
MLX: wire up scheduler selected context size for ps (#16918) |
|
Technical detail and mild tone; no AI ha |
2026-06-26 |
| pr |
0.10 |
mlx: keep loaded model memory resident |
|
Mix of template and technical content, h |
2026-07-24 |
| pr |
0.10 |
cmd/tui: agent system prompt command |
|
Terse, technical, and issue-focused with |
2026-07-21 |
| pr |
0.10 |
agent: permission skill loading |
|
Technical and concise; clear rationale a |
2026-07-22 |
| pr |
0.10 |
test: harden flaky updater and transfer unit tests |
|
Addresses CI/test issues with domain det |
2026-07-24 |
| pr |
0.10 |
cmd: disable word wrap for redirected run output |
|
Clear, specific, technical steps and tes |
2026-06-24 |
| pr |
0.10 |
scripts: install systemd unit to vendor dir so mask/disable |
|
Domain-specific description and user men |
2026-07-24 |
| pr |
0.10 |
openai: propagate ToolName for tool messages with nil conten |
|
Concise bug explanation, domain terms, i |
2026-07-24 |
| pr |
0.10 |
server/download: fix ticker leak in blobDownload.Wait |
|
Brief, focused fix explanation, technica |
2026-07-24 |
| pr |
0.10 |
manifest: fix nil error wrapped with %w in Manifests |
|
Technical bug fix with concise explanati |
2026-07-24 |
| pr |
0.10 |
server: prevent nil CancelFunc panic in blobDownload/Upload |
|
Explains code context and fix concisely, |
2026-07-24 |
| pr |
0.10 |
server/model: fix file descriptor leak in parseFromModel |
|
Focuses on resource leak and its fix, do |
2026-07-24 |
| pr |
0.10 |
fix(openai): accept reasoning_effort minimal mapping to low |
|
Short, uses commit prefix and domain map |
2026-07-20 |
| pr |
0.10 |
fix: prevent sidebar open animation on initial load |
|
Terse, ticket ref, practical commit lang |
2026-07-20 |
| pr |
0.10 |
x/create: quantize lm_head at 8-bit in the requested family |
|
Domain-specific detail, concise language |
2026-07-23 |
| pr |
0.10 |
docs: include model name in ollama create example |
|
Concise, direct, a few minor formalities |
2026-07-23 |
| pr |
0.10 |
docs: include model name in ollama create example |
|
Concise, direct, a few minor formalities |
2026-07-23 |
| pr |
0.10 |
x/imagegen: fix image generation on the CUDA backend (flux2- |
|
Technical detail, complex troubleshootin |
2026-07-22 |
| pr |
0.10 |
server: add metrics endpoint |
|
Casual, domain-specific, with typos and |
2026-07-01 |
| pr |
0.10 |
feat(agent): highlight fenced code blocks |
|
Uses domain-specific language, concise b |
2026-07-21 |
| pr |
0.10 |
model: align Laguna with upstream llama.cpp |
|
Technical summary, brief, some formal to |
2026-07-22 |
| pr |
0.10 |
Improve /prompt rendering performance |
|
Domain-specific, concise, not AI-typical |
2026-07-22 |
| pr |
0.10 |
model/parsers: finalize incomplete GLM tool calls |
|
Domain-specific details and error analys |
2026-07-18 |
| pr |
0.10 |
llm: raise and expose Gemma 4 image token budget |
|
Technical, specific context; informal se |
2026-07-13 |
| pr |
0.10 |
create: add server-side MLX imports and drop GGUF conversion |
|
Contains technical explanations and vali |
2026-03-19 |
| pr |
0.10 |
docs: document reasoning effort values for Codex integration |
|
Well-detailed docs update, but no stereo |
2026-07-22 |
| pr |
0.10 |
docs: document reasoning effort values for Codex integration |
|
Summary uses user perspective and techni |
2026-07-19 |
| pr |
0.10 |
fix: honor Codex App num_ctx metadata |
|
Technical, uses inline test commands, no |
2026-05-30 |
| pr |
0.10 |
docs: add ChatOSS to community integrations |
|
Very brief, domain-specific, matches a s |
2026-07-22 |
| pr |
0.10 |
model: add Laguna MLX support with custom kernels |
|
Succinct summary with context; technical |
2026-07-17 |
| pr |
0.10 |
server: retry interrupted model manifest pulls |
|
Informal, concise, and technical with do |
2026-07-19 |
| pr |
0.10 |
test: revamp integration test entrypoints |
|
Refactoring description, well-written bu |
2026-06-06 |
| pr |
0.10 |
model: add Laguna v8 chat support and fix Metal inference |
|
Detailed but includes domain jargon and |
2026-07-21 |
| pr |
0.10 |
install.sh: detect AMD ROCm on WSL2 (ROCDXG) and enable it o |
|
Detailed technical explanation, clear do |
2026-06-05 |
| pr |
0.10 |
server: truncate embedding inputs to the physical batch size |
|
Technical detail and natural, concise bu |
2026-06-12 |
| pr |
0.10 |
anthropic: serialize empty message content as an empty array |
|
Precise, domain-specific bug fix; inform |
2026-07-21 |
| pr |
0.10 |
llm: account for multiple model memory buffers |
|
Domain-specific, concise, with technical |
2026-07-19 |
| pr |
0.10 |
docs/api: add streaming OpenAI compatibility examples |
|
Template for changelog, domain-relevant, |
2026-07-21 |
| pr |
0.10 |
win: support CUDA on Windows ARM64 |
|
Short, precise technical summary without |
2026-06-26 |
| pr |
0.10 |
tools: return buffered output as content when no tool call i |
|
Detailed technical explanation; no AI ph |
2026-07-21 |
| pr |
0.10 |
docs: add ChatOSS to community integrations |
|
Very brief, domain-specific, matches a s |
2026-07-21 |
| commit |
0.05 |
test: revamp integration test entrpoints (#16560) |
|
Terse, contains domain typos and concise |
2026-07-21 |
| pr |
0.05 |
docs: add model name to ollama create example |
|
Brief, clearly written fix with technica |
2026-07-24 |
| pr |
0.05 |
fix(readline): support Ctrl+Left/Right word navigation |
|
Detailed technical explanation with doma |
2026-06-20 |
| pr |
0.05 |
mlx: implement Nemotron 3 Nano Omni |
|
Technical free-text with abbreviations a |
2026-07-06 |
| pr |
0.05 |
ggml: Use max graph memory allocation when reserving |
|
Technical, concise, with human style and |
2025-11-26 |
| pr |
0.05 |
llm: support multiple LoRA adapters and hot-swapping |
|
Engaged, casual tone with domain referen |
2026-02-02 |
| pr |
0.05 |
parsers: track nesting depth when closing qwen3-coder tool c |
|
Technical, concise, minimal formality, d |
2026-07-03 |
| pr |
0.05 |
Add whatbroke to community integrations |
|
Terse, specific, minor typos, heavy doma |
2026-07-23 |
| pr |
0.05 |
llm: compute TrimSpace once in token repeat detection |
|
Terse, technical changelog with no AI-st |
2026-07-23 |
| pr |
0.05 |
fs/ggml: avoid strings.Reader alloc in writeGGUFString |
|
Technical, with code and benchmarks; no |
2026-07-23 |
| pr |
0.05 |
convert: hoist static regexps to package level |
|
Concise bullet points, domain-specific, |
2026-07-23 |
| pr |
0.05 |
app: coalesce chat stream updates |
|
Casual intro, technical explanation, con |
2026-07-19 |
| commit |
0.00 |
cmd/tui: agent system prompt command (#17296) |
|
Template section only, no human or AI fr |
2026-07-24 |
| commit |
0.00 |
agent: permission skill loading (#17304) |
|
Template section only, no human or AI fr |
2026-07-24 |
| commit |
0.00 |
qwen3_5: gather packed gate_up experts in one launch |
|
Uses jargon, casual tone, some minor typ |
2026-07-22 |
| commit |
0.00 |
qwen3_5: decode each expert tensor with its own quantization |
|
Contains domain jargon, concise; human s |
2026-07-21 |
| commit |
0.00 |
x/create: quantize lm_head at 8-bit in the requested family |
|
Technical, nuanced explanation, informal |
2026-07-23 |
| commit |
0.00 |
agent: import skills from coding agents (#17294) |
|
Very terse commit title; human-written s |
2026-07-23 |
| commit |
0.00 |
model: align Laguna with upstream llama.cpp (#17335) |
|
Technical summary with abbreviations and |
2026-07-23 |
| commit |
0.00 |
docs: update retirements (#17289) |
|
Very short message, typical commit conve |
2026-07-22 |
| commit |
0.00 |
mlx update (#17332) |
|
Minimal placeholder, human style, no AI |
2026-07-22 |
| commit |
0.00 |
CI: fix missing CUDA v13.4 sub-package (#17288) |
|
Concise engineering commit message with |
2026-07-21 |
| commit |
0.00 |
server: detect download stalls before the first byte (#17259 |
|
Terse technical commit, natural for Git |
2026-07-21 |
| commit |
0.00 |
win: support CUDA on Windows ARM64 (#16931) |
|
Short, domain-specific commit; human sty |
2026-07-21 |
| commit |
0.00 |
llama: enable dio on linux CUDA/ROCm iGPUs (#17286) |
|
Brief technical summary, clear engineeri |
2026-07-21 |
| commit |
0.00 |
docs: add renderer/parser fields to the API docs (#17275) |
|
Direct and concise documentation change. |
2026-07-20 |
| commit |
0.00 |
cuda: add CC 10.0 for linux in CUDA v12 (#17025) |
|
Technical, concise, and domain-specific |
2026-07-20 |
| commit |
0.00 |
build: bump Linux toolchain to GCC 13 (#17244) |
|
Brief, technical, and includes domain re |
2026-07-20 |
| commit |
0.00 |
MLX update (#17189) |
|
Extremely terse, minimal content; typica |
2026-07-20 |
| commit |
0.00 |
launch: update Hermes integration (#17202) |
|
Very brief update text, no AI generation |
2026-07-20 |
| commit |
0.00 |
cmd/tui/chat: render bold emphasis consistently across markd |
|
Consistent with human-written, direct te |
2026-07-20 |
| commit |
0.00 |
cmd: remove standalone agent command (#17229) |
|
Succinct and uses standard human commit |
2026-07-20 |
| commit |
0.00 |
cmd: complete slash commands before submitting (#17230) |
|
Short, human style typical for commit me |
2026-07-20 |
| commit |
0.00 |
agent/tui: remove redundant context-window refreshes from ev |
|
Direct and concise; standard human engin |
2026-07-20 |
| commit |
0.00 |
cmd: route root command server start through checkServerHear |
|
Detailed but informal and technical; hum |
2026-07-20 |
| commit |
0.00 |
llama.cpp update (#17186) |
|
Minimal, with only a brief subject; huma |
2026-07-20 |
| commit |
0.00 |
agent: skills system (#17203) |
|
Concise commit message with domain jargo |
2026-07-17 |
| commit |
0.00 |
anthropic: close text block before starting thinking block ( |
|
Brief message with code-specific termino |
2026-07-17 |
| commit |
0.00 |
agent: allow unlimited tool rounds for cloud models by defau |
|
Direct commit message using standard tec |
2026-07-17 |
| commit |
0.00 |
agent: clean up semantics, UX, DX, and procedural code (#172 |
|
Uses standard technical terms, terse and |
2026-07-17 |
| commit |
0.00 |
agent: reorder working directory instruction (#17228) |
|
Very terse, domain-specific commit messa |
2026-07-17 |
| commit |
0.00 |
cmd: remove dead agent prompt wrappers (#17227) |
|
Short, descriptive technical commit mess |
2026-07-17 |
| commit |
0.00 |
launch: keep Claude Code channels available (#17210) |
|
Concise and contextually relevant techni |
2026-07-16 |
| commit |
0.00 |
cmd: put current working dir in the system prompt (#17188) |
|
Terse commit message with domain jargon, |
2026-07-15 |
| commit |
0.00 |
agent/tools: isolate web tests from cloud policy (#17208) |
|
Short, jargon-filled commit message; no |
2026-07-15 |
| commit |
0.00 |
fix launch model picker recovery (#17170) |
|
Plain, brief, typo-like commit summary, |
2026-07-15 |
| commit |
0.00 |
docs: add VS Code extension setup (#17158) |
|
Concise, standard doc commit message, no |
2026-07-15 |
| commit |
0.00 |
docs: collapsed previous retirements (#17167) |
|
Short, template-like commit; no AI hallm |
2026-07-14 |
| commit |
0.00 |
agent/tools: surface actionable web auth error (#17169) |
|
Terse and specific; no AI characteristic |
2026-07-14 |
| commit |
0.00 |
mlxrunner: raise the MTP pending-flush cap to 256 tokens |
|
Technical, informal tone, clear domain k |
2026-07-07 |
| commit |
0.00 |
mlxrunner: key the cache trie by token pairs for draft cache |
|
Technical jargon, concise and informal s |
2026-07-06 |
| commit |
0.00 |
mlxrunner: construct per-model state at load |
|
Technical, includes specifics, not AI-ty |
2026-07-06 |
| commit |
0.00 |
mlxrunner: rename kvCache to prefixCache |
|
Technical rename, brief rationale, no AI |
2026-07-06 |
| commit |
0.00 |
cache: stop recurrent conv state from pinning the forward bu |
|
Technical explanation, uses domain terms |
2026-06-03 |
| commit |
0.00 |
nn/recurrent: derive conv boundary states from a single conv |
|
Technical explanation, natural phrasing, |
2026-06-03 |
| commit |
0.00 |
mlx: configurable model load timeout (#14796) |
|
Template-like PR title, no free-form AI |
2026-07-13 |
| commit |
0.00 |
launch: rename Codex App integration to ChatGPT (#17161) |
|
Standard terse commit with no AI signals |
2026-07-13 |
| commit |
0.00 |
cmd/tui: simplify integration selection and update menu desc |
|
Standard changelog commit, no signs of A |
2026-07-13 |
| commit |
0.00 |
mlx: restore NAX in Metal v4 builds (#17160) |
|
Domain-specific technical explanation, h |
2026-07-13 |
| commit |
0.00 |
cmd: agent UI (#17017) |
|
Very terse commit message, no AI hallmar |
2026-07-10 |
| commit |
0.00 |
launch: warn before launching old agent models (#17063) |
|
Short, domain-specific; likely human-wri |
2026-07-09 |
| commit |
0.00 |
create: select the qwen3.5 parser and renderer for Qwen3.5/N |
|
Technical, concise, uses domain-specific |
2026-06-03 |
| commit |
0.00 |
llm: allow iGPU mmproj offload with fit padding (#16996) |
|
Contains domain-specific jargon and info |
2026-07-07 |
| commit |
0.00 |
server: apply format constraint for all thinking parsers whe |
|
Very terse template-based message; no AI |
2026-07-07 |
| commit |
0.00 |
server: remove unused internal/ code (#17071) |
|
Extremely terse commit message; clearly |
2026-07-07 |
| commit |
0.00 |
create: harden GGUF create flows (#17062) |
|
Terse, uses abbreviation and developer s |
2026-07-06 |
| commit |
0.00 |
launch: disable claude code telemetry by default (#17061) |
|
Short commit title, no AI phrasing or to |
2026-07-06 |
| commit |
0.00 |
mlx: update to de7b4ed9 (#17056) |
|
Standard commit update, no AI patterns e |
2026-07-06 |
| commit |
0.00 |
server: remove OLLAMA_EXPERIMENT=client2 (#16962) |
|
Direct removal note, concise and human-l |
2026-07-06 |
| commit |
0.00 |
llama.cpp update (#17055) |
|
Brief update, typical human committer st |
2026-07-06 |
| commit |
0.00 |
llama: enable FA on CUDA CC 6.x GPUs (#16994) |
|
Technical language with domain specifics |
2026-07-03 |
| commit |
0.00 |
compat: use UTF-8-safe file open (#16999) |
|
Brief, technical summary with human-like |
2026-07-02 |
| commit |
0.00 |
rocm: remove no longer supported devices (#17010) |
|
Informal, human tone and concise problem |
2026-07-02 |
| commit |
0.00 |
llama: clean up dead code from llama-server work (#17007) |
|
Terse, technical language; typical human |
2026-07-02 |
| commit |
0.00 |
agent: harness core (#16963) |
|
Minimal, standard changelog line with no |
2026-07-02 |
| commit |
0.00 |
discover: fall back to standard CUDA when the JetPack runner |
|
Extensive technical detail, domain jargo |
2026-07-02 |
| commit |
0.00 |
docs(cloud): update retirement list (#17000) |
|
Standard doc update, no free-text or AI |
2026-07-02 |
| commit |
0.00 |
fix cuda toolkit lookup and parallel (#16613) |
|
Terse commit messages with technical jar |
2026-06-30 |
| commit |
0.00 |
ci: avoid unbounded parallelism (#16966) |
|
Casual tone, domain-specific explanation |
2026-06-30 |
| commit |
0.00 |
mlx: tighten up gemma4 moe loading code (#16964) |
|
Technical content with human phrasing an |
2026-06-30 |
| commit |
0.00 |
launch: check for min version for hermes desktop (#16912) |
|
Terse commit message, technical, no AI s |
2026-06-29 |
| commit |
0.00 |
llama.cpp update (#16960) |
|
Brief update message, no AI characterist |
2026-06-29 |
| commit |
0.00 |
mlx: bump dependency (#16935) |
|
Contains tech jargon, concise and direct |
2026-06-29 |
| commit |
0.00 |
tools: ignore braces inside JSON strings when detecting tool |
|
Concise, technical language with some in |
2026-06-27 |
| commit |
0.00 |
parser/renderer: add Ornith 9B renderer/parser support (#169 |
|
Very terse, uses minimal language; human |
2026-06-26 |
| commit |
0.00 |
launch: update hermes install urls to official (#16913) |
|
Concise, technical commit, typical human |
2026-06-25 |
| commit |
0.00 |
fix capability grid dark mode style (#16907) |
|
Terse bugfix commit, minimal context, cl |
2026-06-25 |
| commit |
0.00 |
CUDA: require driver 550 or newer for v12 (#16895) |
|
Contains technical jargon, details, huma |
2026-06-25 |
| commit |
0.00 |
docs: redesign coding integration docs (#16808) |
|
Very brief, typical for docs update comm |
2026-06-25 |
| commit |
0.00 |
llama.cpp version update (#16548) |
|
Sparse, standard technical commit title, |
2026-06-24 |
| commit |
0.00 |
jetson: add CC 87 for CUDA v13 (#16628) |
|
Concise title and brief technical explan |
2026-06-24 |
| commit |
0.00 |
server: align generate with native chat templates (#16878) |
|
Technical jargon, informal phrasing, no |
2026-06-24 |
| commit |
0.00 |
docs: redesign docs landing and integrations overview (#1680 |
|
Standard docs commit message with co-aut |
2026-06-24 |
| commit |
0.00 |
llm: fix ollama ps double-counting mmap'd weights on partial |
|
Detailed technical explanation, normal i |
2026-06-24 |
| commit |
0.00 |
mlx: update and fix CUDA JIT packaging (#16871) |
|
Technical, specific, human-like explanat |
2026-06-24 |
| commit |
0.00 |
llm: use host Vulkan loader on Windows (#16869) |
|
Short technical summary, human phrasing. |
2026-06-24 |
| commit |
0.00 |
llama: default qwen2.5vl window attention metadata (#16868) |
|
Detailed domain-specific description, in |
2026-06-24 |
| commit |
0.00 |
llm: preserve generation headroom for shifted prompts (#1685 |
|
Brief message, touches on context, no AI |
2026-06-23 |
| commit |
0.00 |
docs: document max think level (#16877) |
|
Concise, uses common docs commit format, |
2026-06-23 |
| commit |
0.00 |
llm: size mmproj offload by projector memory (#16866) |
|
Technical, domain acronyms, no generic o |
2026-06-23 |
| commit |
0.00 |
llama: add sm_86 architecture to cuda_v13_windows preset (#1 |
|
Detailed technical explanation, signed b |
2026-06-23 |
| commit |
0.00 |
launch/codex: detect model drift when Codex App UI switches |
|
Concise, technical summary with context- |
2026-06-22 |
| commit |
0.00 |
mlxrunner: choose the speculative draft length to maximize t |
|
Uses technical abbreviations and concise |
2026-06-10 |
| pr |
0.00 |
Clean up dead code |
|
Extremely terse and informal, lacks AI s |
2026-07-24 |
| pr |
0.00 |
qwen3_5: fix expert quantization handling and gather packed |
|
Dense technical language, code issues, a |
2026-07-22 |
| pr |
0.00 |
server: fix ps data race on scheduler loaded map |
|
Technically detailed and focused on conc |
2026-07-24 |
| pr |
0.00 |
docs: add community integration to README |
|
Clearly human, conversational and inform |
2026-03-24 |
| pr |
0.00 |
model: add Laguna MLX support |
|
Contains domain-specific vocabulary and |
2026-07-17 |
| pr |
0.00 |
server: cancel pull on client disconnect |
|
List format, informal and technical lang |
2026-04-29 |
| pr |
0.00 |
model/parsers: tolerate qwen tool-call format drift (qwen3.6 |
|
Technical, contains domain-specific deta |
2026-06-01 |
| pr |
0.00 |
model/parsers: accept bare qwen tool calls |
|
Detailed, technical, uses abbreviations |
2026-07-23 |
| pr |
0.00 |
llm: use direct I/O for Vulkan iGPUs |
|
Contains domain-specific detail and info |
2026-07-23 |
| pr |
0.00 |
Regression in 0.32.1: /api/generate aborts with token repeat |
|
Highly technical with code references an |
2026-07-23 |
| pr |
0.00 |
ollamarunner: Automatically enable flash attention |
|
Technical, covers advanced logic, no AI |
2025-12-12 |
| pr |
0.00 |
docs: update retirements |
|
Very terse commit, no AI hallmarks or fr |
2026-07-21 |
| pr |
0.00 |
mlx update |
|
Very brief, no AI signals. |
2026-07-22 |
| pr |
0.00 |
cmd: create SSH keypair with O_EXCL to avoid overwrite race |
|
Uses domain jargon and informal phrasing |
2026-07-22 |
| pr |
0.00 |
server/llm: stop parsing special-token literals inside messa |
|
Technical analysis, direct problem descr |
2026-07-20 |
| pr |
0.00 |
launch: update glm-5.2:cloud metadata and OMP reasoning flag |
|
Brief technical summary with no AI-style |
2026-07-22 |
| pr |
0.00 |
middleware: support allowed_domains and blocked_domains for |
|
Terse technical description, uses abbrev |
2026-07-22 |
| pr |
0.00 |
server: warm prefill cache across model unload/reload |
|
Technical content, detailed test plan, d |
2026-07-21 |
| pr |
0.00 |
docs(zh-TW): add Traditional Chinese README and quickstart g |
|
Direct writing, mentions language specif |
2026-07-22 |
| pr |
0.00 |
build: forward explicit CUDAToolkit_ROOT to CUDA sub-builds |
|
Technical PR, detailed explanation, no A |
2026-07-22 |
| pr |
0.00 |
Add model eval metrics to /metrics |
|
Direct, example-driven, markdown snippet |
2025-06-22 |
| pr |
0.00 |
CI: fix missing CUDA v13.4 sub-package |
|
Single line, context-specific, concise h |
2026-07-21 |
| pr |
0.00 |
server: detect download stalls before the first byte |
|
Detailed, informal technical explanation |
2026-07-19 |
| pr |
0.00 |
llama: enable dio on linux CUDA/ROCm iGPUs |
|
Concise technical summary with domain-sp |
2026-07-21 |
| pr |
0.00 |
fix: report actual VRAM available for AMD integrated GPUs (A |
|
Detailed technical analysis, domain abbr |
2026-07-21 |
| pr |
0.00 |
add renderer/parser fields to the API docs |
|
Commit message is minimal and uses domai |
2026-07-20 |
| pr |
0.00 |
docs: clarify Claude Code context limits |
|
Changelog with specific items and human |
2026-07-06 |
| pr |
0.00 |
cuda: add CC 10.0 for linux in CUDA v12 |
|
Technical summary with domain references |
2026-07-04 |
| pr |
0.00 |
docs: update retirements |
|
Very terse commit, no AI hallmarks or fr |
2026-07-20 |
