* create: Clean up experimental paths
This cleans up the experimental features, and adds both unit and integration test coverage to verify no regressions.
* create: preserve config and layer names when creating from safetensors models
When creating a model FROM an existing safetensors model, ModelFormat,
Capabilities, and layer Name fields were lost. ModelFormat stayed empty
because it's only set from GGML layers (which safetensors models lack),
and layer names weren't copied in parseFromModel. This caused derived
models to fail loading ("config.json not found in manifest").
* review comments
* mlx: Improve M5 performance with NAX
This modifies the Mac release to now have 2 builds of MLX for broader
compatibility while supporting the latest M5 hardware features. NAX requires
building with xcode 26.2 and targetting support only for OS v26 and up. Since
we want to support older MacOS versions as well, we now need 2 different MLX
builds and runtime detection logic to select the optimal version. The newer
build will detect NAX missing at runtime, so it is safe to run on pre M5 macs.
* mac: prevent generate on cross-compiles
For some versions of Xcode, cmake builds are failing due to header problems in
cross-compiling during the generate phase. Since generate is producing arch
independent generated output, we can skip this during cross-compiling.
Replace hardcoded Encode(prompt, true) with
Encode(prompt, r.Tokenizer.AddBOS()) so the pipeline respects each
model's tokenizer configuration.
Models with add_bos_token=true (gemma3, llama): unchanged, tokenizer
still prepends BOS.
Models with bos_token=null (qwen3, qwen3.5): unchanged, the BOS
guard (vocab.BOS >= 0) already prevented prepending regardless of
the flag.
This aligns the pipeline with the /v1/tokenize endpoint which already
uses Tokenizer.AddBOS().
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.
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.
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.
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
* 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.
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.
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.
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.
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.
Add QuantizedEmbedding and EmbeddingLayer interface so models can
use quantized embedding weights and expose tied output projections.
This change updates gemma3, glm4_moe_lite, llama, qwen3, and qwen3_5
to use the new interface.
This change adds a tensorImportTransform interface for model-specific
tensor transformations during safetensors import. This allows importing
and modifying the standard HF based weights as well as the mlx-community
derived pre-quantized safetensors repos to be directly
imported into `ollama create`. Right now this only works with Qwen3.5
importing which does tensor renaming, norm weight shifting (it
adds +1 to each value of the norm vectors), conv1d transposition,
and casts to BF16s for F32 based vectors.
MLX runners (image generation and LLM) previously bypassed the
scheduler's standard load path via a separate loadMLX method. This meant
they skipped VRAM fitting checks and couldn't participate in model
eviction.
Now all model types flow through the same load function. Model eviction
for MLX is based on weights as KV cache and compute graph are dynamic.
This means that eviction does not take into account the worst case
memory and models can still compete for memory but it is a significant
improvement.
The CLI now links to the lazy-load MLX code, but that still happens in
init functions. On internal MLX errors, the CLI exits before it has a
chance to start. This change re-wires the MLX error handling so it
doesn't exit by default. The MLX based runners currently expect exits
on failure, so they re-initialize the default error handling. We can
refine error handling for better go stack traces in the future.
* prefer rocm v6 on windows
Avoid building with v7 - more changes are needed
* MLX: add header vendoring and remove go build tag
This switches to using a vendoring approach for the mlx-c headers so that Go
can build without requiring a cmake first. This enables building the new MLX
based code by default. Every time cmake runs, the headers are refreshed, so we
can easily keep them in sync when we bump mlx versions. Basic Windows
and Linux support are verified.
* ci: harden for flaky choco repo servers
CI sometimes fails due to choco not actually installing cache. Since it just speeds up the build, we can proceed without.
* review comments
- Collapse MLX sampling state into a single sample.Sampler struct (options + history).
- Replace interface-based sampler chain (TopP, TopK, penalty, etc.) with function-based transforms.
- Update request/pipeline wiring to use *sample.Sampler, seed history from prompt tokens, and append generated tokens each step.
- Implement top_p, min_p, repeat_penalty, and frequency_penalty
GLM models sometimes omits </arg_value> closing tags in tool call XML, causing xml.Unmarshal to fail with "element <arg_value> closed by </tool_call>".
This is a known issue across the GLM family.
Sanitize the input to fix closing arg_key values so encoding/xml can handle it.
This change adds support for qwen3.5-next-moe models (qwen3-next/qwen3.5-next/qwen3-coder) to the MLX runner. It also:
* introduces recurrent cache support and related MLX ops
* updates pipeline/runner integration and adds tests
* properly quantizes stacked expert tensors
* a Gated Delta Metal kernel for fast SSM inference
* adds new MLX calls for Conv1d, DepthwideConv1d, Contiguous, Exp, Log, SoftmaxAxis
* don't require pulling stubs for cloud models
This is a first in a series of PRs that will better integrate Ollama's
cloud into the API and CLI. Previously we used to have a layer of
indirection where you'd first have to pull a "stub" model that contains
a reference to a cloud model. With this change, you don't have to pull
first, you can just use a cloud model in various routes like `/api/chat`
and `/api/show`. This change respects
<https://github.com/ollama/ollama/pull/14221>, so if cloud is disabled,
these models won't be accessible.
There's also a new, simpler pass-through proxy that doesn't convert the
requests ahead of hitting the cloud models, which they themselves
already support various formats (e.g., `v1/chat/completions` or Open
Responses, etc.). This will help prevent issues caused by double
converting (e.g., `v1/chat/completions` converted to `api/chat` on the
client, then calling cloud and converting back to a
`v1/chat/completions` response instead of the cloud model handling the
original `v1/chat/completions` request first).
There's now a notion of "source tags", which can be mixed with existing
tags. So instead of having different formats like`gpt-oss:20b-cloud` vs.
`kimi-k2.5:cloud` (`-cloud` suffix vs. `:cloud`), you can now specify
cloud by simply appending `:cloud`. This PR doesn't change model
resolution yet, but sets us up to allow for things like omitting the
non-source tag, which would make something like `ollama run
gpt-oss:cloud` work the same way that `ollama run gpt-oss` already works
today.
More detailed changes:
- Added a shared model selector parser in `types/modelselector`:
- supports `:cloud` and `:local`
- accepts source tags in any position
- supports legacy `:<tag>-cloud`
- rejects conflicting source tags
- Integrated selector handling across server inference/show routes:
- `GenerateHandler`, `ChatHandler`, `EmbedHandler`,
`EmbeddingsHandler`, `ShowHandler`
- Added explicit-cloud passthrough proxy for ollama.com:
- same-endpoint forwarding for `/api/*`, `/v1/*`, and `/v1/messages`
- normalizes `model` (and `name` for `/api/show`) before forwarding
- forwards request headers except hop-by-hop/proxy-managed headers
- uses bounded response-header timeout
- handles auth failures in a friendly way
- Preserved cloud-disable behavior (`OLLAMA_NO_CLOUD`)
- Updated create flow to support `FROM ...:cloud` model sources (though
this flow uses the legacy proxy still, supporting Modelfile overrides
is more complicated with the direct proxy approach)
- Updated CLI/TUI/config cloud detection to use shared selector logic
- Updated CLI preflight behavior so explicit cloud requests do not
auto-pull local stubs
What's next?
- Cloud discovery/listing and cache-backed `ollama ls` / `/api/tags`
- Modelfile overlay support for virtual cloud models on OpenAI/Anthropic
request families
- Recommender/default-selection behavior for ambiguous model families
- Fully remove the legacy flow
Fixes: https://github.com/ollama/ollama/issues/13801
* consolidate pull logic into confirmAndPull helper
pullIfNeeded and ShowOrPull shared identical confirm-and-pull logic.
Extract confirmAndPull to eliminate the duplication.
* skip local existence checks for cloud models
ModelExists and the TUI's modelExists both check the local model list,
which causes cloud models to appear missing. Return true early for
explicit cloud models so the TUI displays them beside the integration
name and skips re-prompting the model picker on relaunch.
* support optionally pulling stubs for newly-style names
We now normalize names like `<family>:<size>:cloud` into legacy-style
names like `<family>:<size>-cloud` for pulling and deleting (this also
supports stripping `:local`). Support for pulling cloud models is
temporary, once we integrate properly into `/api/tags` we won't need
this anymore.
* Fix server alias syncing
* Update cmd/cmd.go
Co-authored-by: Parth Sareen <parth.sareen@ollama.com>
* address comments
* improve some naming
---------
Co-authored-by: ParthSareen <parth.sareen@ollama.com>
Only the last token's processing time is included in prompt processing,
giving an artificially high rate. In addition, the number of tokens
only included the tokens that miss the cache, instead of our historic
total tokens.
Currently, context length is unbounded - the cache will keep
growing forever independent of the model's trained context
length. This caps it and enforces semantics similar to most
cloud services:
- Long prompts will result in an error, not truncation.
- Generation that exceeds the context will be stopped
Errors that occur during pipeline processing are currently only
logged but not sent back to the client. Rather than using HTTP
status codes as we have historically done, this serializes errors
as messages to allow sending them at any time during the stream.
The MLX runner previously reported a static VRAM estimate that was
computed at load time and consisted only of the weights. This is
strictly less than the actual memory usage, as it does not include
the KV cache or compute graph.
When the entire prompt was already cached (e.g. repeated prompt),
findRemaining returned an empty slice, causing FromValues to panic
on an index-out-of-range accessing a zero-length byte slice.
Fix by always keeping at least one token to re-evaluate so the
pipeline can seed token generation. Also reject empty prompts
early rather than panicking.
Currently, a canceled request can result in computation continuing
in the background to completion. It can also trigger a deadlock
when there is nobody to read the output tokens and the pipeline
cannot continue to the next request.
Particularly in error cases, it can be difficult to ensure that
all pinned memory is unpinned, MLX buffers are released and cache
state is consistent. This encapsulates those pieces and sets up
proper deferrals so that this happens automatically on exit.
