5246 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
Jeffrey Morgan	31f968fe1f	cmd: set OpenCode default model in config (#15127) v0.19.0 v0.19.0-rc2	2026-03-29 12:11:36 -07:00
Jeffrey Morgan	b7bda92d52	model: add qwen3-next compatibility for legacy ssm_in projections (#15133)	2026-03-29 11:50:47 -07:00
Parth Sareen	8e54823fd3	revert context length warnings change (#15121)	2026-03-28 16:43:59 -07:00
Parth Sareen	7c8da5679e	launch: improve multi-select for already added models (#15113)	2026-03-28 13:44:40 -07:00
Parth Sareen	6214103e66	launch: auto-install pi and manage web-search lifecycle (#15118)	2026-03-28 13:06:20 -07:00
Patrick Devine	9e7cb9697e	mlx: fix vision capability + min version (#15106) v0.19.0-rc1	2026-03-27 17:09:28 -07:00
Bruce MacDonald	3824e380a8	server: preserve raw manifest bytes during pull (#15104) ... pullModelManifest unmarshals the registry response into a Go struct then re-marshals with json.Marshal before writing to disk. When the registry's JSON formatting or field ordering differs from Go's output, the local SHA256 won't match the registry's Ollama-Content-Digest header, causing false "out of date" warnings. Preserve the raw bytes from the registry response and write them directly to disk so the local manifest is byte-for-byte identical to what the registry serves.	2026-03-27 15:42:31 -07:00
Devon Rifkin	c9b2dcfc52	anthropic: fix empty inputs in content blocks (#15105) ... * anthropic: fix empty inputs in content blocks When we switched to `api.ToolCallFunctionArguments`, `omitempty` stopped doing what we were relying on it for before. This would cause non-tool content blocks to have an `"input": {}` field, which doesn't match our old behavior. * use omitzero instead	2026-03-27 15:41:27 -07:00
Parth Sareen	b00bd1dfd4	launch: skip context length warning for MLX models and show model name (#15102)	2026-03-27 15:01:33 -07:00
Jesse Gross	ac83ac20c4	anthropic: fix KV cache reuse degraded by tool call argument reordering ... Use typed structs for tool call arguments instead of map[string]any to preserve JSON key order, which Go maps do not guarantee.	2026-03-27 14:30:16 -07:00
Bruce MacDonald	e7ccc129ea	app: fix false "out of date" model warnings (#15101) ... The staleness check compared the local manifest digest (SHA256 of the file on disk) against the registry's Ollama-Content-Digest header. These never matched because PullModel re-serializes the manifest JSON before writing, producing different bytes than the registry's original. The fallback comparison (local modified_at vs upstream push time) was also broken: the generated TypeScript Time class discards the actual timestamp value, so Date parsing always produced NaN. Fix by moving the staleness comparison server-side where we have reliable access to both the local manifest file mtime and the upstream push time. The /api/v1/model/upstream endpoint now returns a simple `stale` boolean instead of raw digests for the frontend to compare. Also adds User-Agent to the CORS allowed headers for dev mode.	2026-03-27 14:15:10 -07:00
Jeffrey Morgan	69ed0c2729	parsers: qwen3.5 streaming tool-call parsing and add regression test (#15098)	2026-03-27 14:04:14 -07:00
Alfredo Matas	1cefa749aa	model/parsers: close think block if tool block starts in Qwen3.5 (#15022)	2026-03-27 11:28:34 -07:00
Daniel Hiltgen	aec2fef95d	ci: harden cuda include path handling (#15093) ... On windows we can get multiple include dirs, so find where the headers are then copy from that location. v0.19.0-rc0	2026-03-27 07:57:07 -07:00
Eva H	366625a831	launch: warn when server context length is below 64k for local models (#15044) ... A stop-gap for now to guide users better. We'll add more in-depth recommendations per integration as well. --------- Co-authored-by: Parth Sareen <parth.sareen@ollama.com> v0.18.4-rc1	2026-03-27 00:15:53 -07:00
Daniel Hiltgen	516ebd8548	ci: include mlx jit headers on linux (#15083) ... * ci: include mlx jit headers on linux * handle CUDA JIT headers	2026-03-26 23:10:07 -07:00
Parth Sareen	f567abc63f	tui: update chat title (#15082)	2026-03-26 18:06:53 -07:00
Eva H	1adfc27f04	launch/vscode: prefer known vs code paths over code on PATH (#15073)	2026-03-26 18:06:28 -04:00
Parth Sareen	4a2b9f9dbc	launch: hide cline integration (#15080)	2026-03-26 14:33:43 -07:00
Parth Sareen	e46b67a6cc	launch: hide vs code (#15076) v0.18.4-rc0	2026-03-26 13:52:50 -07:00
Eva H	c000afe76c	doc: update vscode doc (#15064) ... --------- Co-authored-by: ParthSareen <parth.sareen@ollama.com>	2026-03-26 13:45:48 -07:00
Jesse Gross	9d7b18f81e	mlxrunner: combine setStateRaw and setStateDetached into setState	2026-03-26 13:32:11 -07:00
Jesse Gross	4f5999fd3f	mlxrunner: schedule periodic snapshots during prefill ... Add periodic snapshots every 8k tokens and near the end of the prompt so that long prompts can be partially restored and thinking/generation can be retried without full reprocessing.	2026-03-26 13:32:11 -07:00
Jesse Gross	ac5f0dbb6a	mlxrunner: improve eviction and LRU tracking ... Update LRU last used time just on the nodes that actually used during processing rather than all snapshots along the path. This allows eviction to remove nodes more accurately so we can avoid other heuristics to auto-merge nodes.	2026-03-26 13:32:11 -07:00
Jesse Gross	d1151e18a1	mlx: fix KV cache snapshot memory leak ... mlx.Copy shares the backing buffer with its source (via copy_shared_buffer) rather than allocating independent storage. When used to snapshot a slice of the KV cache, the snapshot array holds the entire original cache buffer alive through the shared data pointer — even after eval detaches the computation graph. Replace Copy with Contiguous in Snapshot and Split. Contiguous allocates a compact buffer when the source buffer is significantly larger than the logical slice (Contiguous::eval checks buffer_size > nbytes + 16384), which is always the case for KV cache slices.	2026-03-25 17:26:34 -07:00
rick	ebbce136c7	ggml: force flash attention off for grok	2026-03-25 16:15:49 -07:00
Devon Rifkin	26b9f53f8e	api/show: overwrite basename for copilot chat (#15062) ... Copilot Chat prefers to use `general.basename` in the built-in Ollama integration, but this name isn't usually shown directly to users (and there may be many models that share this name). Instead we pass back `req.Model`, which for this extension is the value that we return from `/api/tags` v0.18.3 v0.18.3-rc2	2026-03-25 14:02:22 -07:00
Eva H	7575438366	cmd: ollama launch vscode (#15060) ... Co-authored-by: Parth Sareen <parth.sareen@ollama.com>	2026-03-25 16:37:02 -04:00
Eva H	7d7c90d702	tui: add left arrow back navigation in model selector (#14940)	2026-03-25 11:53:48 -07:00
Daniel Hiltgen	4fda69809a	ci: fix windows cgo compiler error (#15046) v0.18.3-rc1	2026-03-24 16:45:36 -07:00
Daniel Hiltgen	c9b5da6b0c	integration: improve ability to test individual models (#14948) ... * integration: improve ability to test individual models Add OLLAMA_TEST_MODEL env var to run integration tests against a single model. Enhance vision tests: multi-turn chat with cached image tokens, object counting, spatial reasoning, detail recognition, scene understanding, OCR, and multi-image comparison. Add tool calling stress tests with complex agent-style prompts, large system messages, and multi-turn tool response handling. * review comments	2026-03-24 14:28:23 -07:00
Patrick Devine	de5cb7311f	mlx: add mxfp4/mxfp8/nvfp4 importing (#15015) ... This change allows importing bf16 and converting to mxfp4/mxfp8/nvfp4 and also importing fp8 and converting directly to mxfp8. v0.18.3-rc0	2026-03-24 13:45:44 -07:00
Jesse Gross	95ee7fbd29	mlxrunner: panic on double unpin	2026-03-23 17:44:19 -07:00
Jesse Gross	ec55536734	mlxrunner: show time since last used in cache dump tree	2026-03-23 17:44:19 -07:00
Jesse Gross	77491439c2	mlxrunner: support partial match on pure transformer caches ... Previously, a partial match within a node's edge would truncate the path to the parent snapshot - effectively making all cache types behave as recurrent caches. Caches with only transformer layers can rewind to arbitrary boundary so this restores this capability to improve cache hits	2026-03-23 17:44:19 -07:00
Parth Sareen	b166b36cd2	docs: update Claude Code with Telegram guide (#15026)	2026-03-23 16:31:21 -07:00
Daniel Hiltgen	c2b0bb7a52	mlx: update as of 3/23 (#14789) ... * mlx: update to HEAD on 3/23 Also fixes a few misc vendoring bugs uncovered with this first update. This also renames the version files to make them clearer. * CUDA Fast Gated Delta kernel * mlx: detect eval errors and panic On model errors or missing kernels, don't mask the error, bubble it up.	2026-03-23 11:28:44 -07:00
Bruce MacDonald	22c2bdbd8a	docs: nemoclaw integration (#14962) ... --------- Co-authored-by: ParthSareen <parth.sareen@ollama.com>	2026-03-20 15:27:37 -07:00
Bruce MacDonald	6df6d097d9	launch: skip openclaw gateway health check when no daemon install (#14984)	2026-03-20 15:20:14 -07:00
Jesse Gross	d7c176ab91	llm, mlxrunner: fix done channel value consumed by first receiver ... Receiving from a buffered chan error consumes the value, so only the first caller (WaitUntilRunning, HasExited, or Close) sees the signal. Subsequent receivers block or take the wrong branch. Replace with a closed chan struct{} which can be received from any number of times, and store the error in a separate field.	2026-03-19 17:44:28 -07:00
Jesse Gross	0ff7d724ff	mlx: fix subprocess log deadlock ... The stderr reader used bufio.Scanner which has a 64KB max line size. If the subprocess wrote a line exceeding this limit, the scanner would stop reading, the OS pipe buffer would fill, and the subprocess would deadlock. Replace the scanner with a statusWriter that wraps io.Copy. The writer forwards all stderr to os.Stderr while capturing the last short line (≤256 bytes) for error reporting, avoiding both the deadlock and the need to buffer arbitrarily long lines.	2026-03-19 17:44:28 -07:00
Devon Rifkin	46cb7795e1	add ability to turn on debug request logging (#14106) ... If `OLLAMA_DEBUG_LOG_REQUESTS` is set, then on server startup a temp folder will be created. Upon any inference request, the body will be logged to a file in this folder, as well as a small shell script to "replay" the request using cURL. This is just intended for debugging scenarios, not as something to turn on normally.	2026-03-19 17:08:17 -07:00
Bruce MacDonald	126d8db7f3	parsers: robust xml tool repair (#14961) ... Previous xml repair for glm was a good start, but we need to go further and repair any incorrect open or closing tags Co-authored-by: Dongluo Chen <dongluo.chen@gmail.com>	2026-03-19 11:24:48 -07:00
Eva H	3f3a24b418	app: fix desktop app stuck loading when OLLAMA_HOST is an unspecified bind address (#14885)	2026-03-19 12:57:57 -04:00
Jesse Gross	96e36c0d90	mlxrunner: share KV cache across conversations with common prefixes ... Enable multiple conversations to reuse cached computations when they share token prefixes (e.g. the same system prompt). A prefix trie tracks shared regions so switching between conversations only recomputes tokens that diverge. Inactive conversation state is paged from active GPU memory to other memory and restored on demand, with LRU eviction to keep memory usage bounded.	2026-03-18 16:06:33 -07:00
Jesse Gross	6f8ddbb26b	mlxrunner: fix Slice(0, 0) returning full dimension instead of empty ... Slice used cmp.Or to resolve a zero stop value to the dimension size, intended to support open-ended slices like a[i:]. This made Slice(0, 0) indistinguishable from Slice(), so any slice with a zero stop would silently include the entire dimension instead of being empty. Replace cmp.Or with an explicit End sentinel and resolve negative indices against the dimension size, matching Python/PyTorch semantics.	2026-03-18 16:06:33 -07:00
Eva H	b5e7888414	cmd/launch: skip redundant config writes when model unchanged (#14941)	2026-03-18 17:36:52 -04:00
Parth Sareen	eab4d22269	docs: update claude code and openclaw for web search (#14922)	2026-03-18 14:18:49 -07:00
Bruce MacDonald	5759c2d2d2	launch: fix openclaw not picking up newly selected model (#14943) ... Sessions with a stale model field were not updated when the primary changed, so the old model continued to be used. v0.18.2 v0.18.2-rc1	2026-03-18 13:20:10 -07:00
Bruce MacDonald	42b1c2642b	docs: update minimax-m2.5 references to m2.7 (#14942)	2026-03-18 12:59:28 -07:00