Compare commits

..

2 Commits

Author SHA1 Message Date
Parth Sareen
6b2abfb433 server: add tests and fix isHuggingFaceURL edge case
- Add comprehensive tests for isHuggingFaceURL and getNumDownloadParts
- Fix bug where domains ending in huggingface.co (like nothuggingface.co)
  would incorrectly match as HuggingFace URLs
- Improve code comments with more detailed documentation
2026-01-18 16:45:17 -08:00
Parth Sareen
805ed4644c server: reduce download concurrency for HuggingFace URLs
Reduces concurrent download parts from 16 to 4 for HuggingFace URLs
to avoid triggering rate limits (HTTP 429 errors).

Adds OLLAMA_HF_CONCURRENCY environment variable for users who want
to customize the concurrency level.

Fixes #13297
2026-01-18 16:38:49 -08:00
7 changed files with 255 additions and 25 deletions

View File

@@ -190,7 +190,7 @@ if(MLX_ENGINE)
install(TARGETS mlx mlxc
RUNTIME_DEPENDENCIES
DIRECTORIES ${CUDAToolkit_BIN_DIR} ${CUDAToolkit_BIN_DIR}/x64 ${CUDAToolkit_LIBRARY_DIR}
PRE_INCLUDE_REGEXES cublas cublasLt cudart nvrtc nvrtc-builtins cudnn nccl openblas gfortran
PRE_INCLUDE_REGEXES cublas cublasLt cudart nvrtc cudnn nccl
PRE_EXCLUDE_REGEXES ".*"
RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX

View File

@@ -110,7 +110,7 @@ More Ollama [Python example](https://github.com/ollama/ollama-python/blob/main/e
import { Ollama } from "ollama";
const client = new Ollama();
const results = await client.webSearch("what is ollama?");
const results = await client.webSearch({ query: "what is ollama?" });
console.log(JSON.stringify(results, null, 2));
```
@@ -213,7 +213,7 @@ models](https://ollama.com/models)\n\nAvailable for macOS, Windows, and Linux',
import { Ollama } from "ollama";
const client = new Ollama();
const fetchResult = await client.webFetch("https://ollama.com");
const fetchResult = await client.webFetch({ url: "https://ollama.com" });
console.log(JSON.stringify(fetchResult, null, 2));
```

View File

@@ -22,7 +22,7 @@ Please refer to the [GPU docs](./gpu).
## How can I specify the context window size?
By default, Ollama uses a context window size of 4096 tokens.
By default, Ollama uses a context window size of 2048 tokens.
This can be overridden with the `OLLAMA_CONTEXT_LENGTH` environment variable. For example, to set the default context window to 8K, use:

View File

@@ -1,5 +1,5 @@
---
title: Linux
title: "Linux"
---
## Install
@@ -13,15 +13,14 @@ curl -fsSL https://ollama.com/install.sh | sh
## Manual install
<Note>
If you are upgrading from a prior version, you should remove the old libraries
with `sudo rm -rf /usr/lib/ollama` first.
If you are upgrading from a prior version, you should remove the old libraries with `sudo rm -rf /usr/lib/ollama` first.
</Note>
Download and extract the package:
```shell
curl -fsSL https://ollama.com/download/ollama-linux-amd64.tar.zst \
| sudo tar x -C /usr
curl -fsSL https://ollama.com/download/ollama-linux-amd64.tgz \
| sudo tar zx -C /usr
```
Start Ollama:
@@ -41,8 +40,8 @@ ollama -v
If you have an AMD GPU, also download and extract the additional ROCm package:
```shell
curl -fsSL https://ollama.com/download/ollama-linux-amd64-rocm.tar.zst \
| sudo tar x -C /usr
curl -fsSL https://ollama.com/download/ollama-linux-amd64-rocm.tgz \
| sudo tar zx -C /usr
```
### ARM64 install
@@ -50,8 +49,8 @@ curl -fsSL https://ollama.com/download/ollama-linux-amd64-rocm.tar.zst \
Download and extract the ARM64-specific package:
```shell
curl -fsSL https://ollama.com/download/ollama-linux-arm64.tar.zst \
| sudo tar x -C /usr
curl -fsSL https://ollama.com/download/ollama-linux-arm64.tgz \
| sudo tar zx -C /usr
```
### Adding Ollama as a startup service (recommended)
@@ -113,11 +112,7 @@ sudo systemctl status ollama
```
<Note>
While AMD has contributed the `amdgpu` driver upstream to the official linux
kernel source, the version is older and may not support all ROCm features. We
recommend you install the latest driver from
https://www.amd.com/en/support/linux-drivers for best support of your Radeon
GPU.
While AMD has contributed the `amdgpu` driver upstream to the official linux kernel source, the version is older and may not support all ROCm features. We recommend you install the latest driver from https://www.amd.com/en/support/linux-drivers for best support of your Radeon GPU.
</Note>
## Customizing
@@ -146,8 +141,8 @@ curl -fsSL https://ollama.com/install.sh | sh
Or by re-downloading Ollama:
```shell
curl -fsSL https://ollama.com/download/ollama-linux-amd64.tar.zst \
| sudo tar x -C /usr
curl -fsSL https://ollama.com/download/ollama-linux-amd64.tgz \
| sudo tar zx -C /usr
```
## Installing specific versions
@@ -196,4 +191,4 @@ Remove the downloaded models and Ollama service user and group:
sudo userdel ollama
sudo groupdel ollama
sudo rm -r /usr/share/ollama
```
```

View File

@@ -179,7 +179,7 @@ _build_macapp() {
fi
rm -f dist/Ollama-darwin.zip
ditto -c -k --norsrc --keepParent dist/Ollama.app dist/Ollama-darwin.zip
ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
(cd dist/Ollama.app/Contents/Resources/; tar -cf - ollama ollama-mlx *.so *.dylib *.metallib 2>/dev/null) | gzip -9vc > dist/ollama-darwin.tgz
# Notarize and Staple
@@ -187,7 +187,7 @@ _build_macapp() {
$(xcrun -f notarytool) submit dist/Ollama-darwin.zip --wait --timeout 20m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
rm -f dist/Ollama-darwin.zip
$(xcrun -f stapler) staple dist/Ollama.app
ditto -c -k --norsrc --keepParent dist/Ollama.app dist/Ollama-darwin.zip
ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
rm -f dist/Ollama.dmg

View File

@@ -95,11 +95,48 @@ func (p *blobDownloadPart) UnmarshalJSON(b []byte) error {
}
const (
numDownloadParts = 16
// numDownloadParts is the default number of concurrent download parts for standard downloads
numDownloadParts = 16
// numHFDownloadParts is the reduced number of concurrent download parts for HuggingFace
// downloads to avoid triggering rate limits (HTTP 429 errors). See GitHub issue #13297.
numHFDownloadParts = 4
minDownloadPartSize int64 = 100 * format.MegaByte
maxDownloadPartSize int64 = 1000 * format.MegaByte
)
// isHuggingFaceURL returns true if the URL is from a HuggingFace domain.
// This includes:
// - huggingface.co (main domain)
// - *.huggingface.co (subdomains like cdn-lfs.huggingface.co)
// - hf.co (shortlink domain)
// - *.hf.co (CDN domains like cdn-lfs.hf.co, cdn-lfs3.hf.co)
func isHuggingFaceURL(u *url.URL) bool {
if u == nil {
return false
}
host := strings.ToLower(u.Hostname())
return host == "huggingface.co" ||
strings.HasSuffix(host, ".huggingface.co") ||
host == "hf.co" ||
strings.HasSuffix(host, ".hf.co")
}
// getNumDownloadParts returns the number of concurrent download parts to use
// for the given URL. HuggingFace URLs use reduced concurrency (default 4) to
// avoid triggering rate limits. This can be overridden via the OLLAMA_HF_CONCURRENCY
// environment variable. For non-HuggingFace URLs, returns the standard concurrency (16).
func getNumDownloadParts(u *url.URL) int {
if isHuggingFaceURL(u) {
if v := os.Getenv("OLLAMA_HF_CONCURRENCY"); v != "" {
if n, err := strconv.Atoi(v); err == nil && n > 0 {
return n
}
}
return numHFDownloadParts
}
return numDownloadParts
}
func (p *blobDownloadPart) Name() string {
return strings.Join([]string{
p.blobDownload.Name, "partial", strconv.Itoa(p.N),
@@ -271,7 +308,11 @@ func (b *blobDownload) run(ctx context.Context, requestURL *url.URL, opts *regis
}
g, inner := errgroup.WithContext(ctx)
g.SetLimit(numDownloadParts)
concurrency := getNumDownloadParts(directURL)
if concurrency != numDownloadParts {
slog.Info(fmt.Sprintf("using reduced concurrency (%d) for HuggingFace download", concurrency))
}
g.SetLimit(concurrency)
for i := range b.Parts {
part := b.Parts[i]
if part.Completed.Load() == part.Size {

194
server/download_test.go Normal file
View File

@@ -0,0 +1,194 @@
package server
import (
"net/url"
"testing"
"github.com/stretchr/testify/assert"
)
func TestIsHuggingFaceURL(t *testing.T) {
tests := []struct {
name string
url string
expected bool
}{
{
name: "nil url",
url: "",
expected: false,
},
{
name: "huggingface.co main domain",
url: "https://huggingface.co/some/model",
expected: true,
},
{
name: "cdn-lfs.huggingface.co subdomain",
url: "https://cdn-lfs.huggingface.co/repos/abc/123",
expected: true,
},
{
name: "cdn-lfs3.hf.co CDN domain",
url: "https://cdn-lfs3.hf.co/repos/abc/123",
expected: true,
},
{
name: "hf.co shortlink domain",
url: "https://hf.co/model",
expected: true,
},
{
name: "uppercase HuggingFace domain",
url: "https://HUGGINGFACE.CO/model",
expected: true,
},
{
name: "mixed case HF domain",
url: "https://Cdn-Lfs.HF.Co/repos",
expected: true,
},
{
name: "ollama registry",
url: "https://registry.ollama.ai/v2/library/llama3",
expected: false,
},
{
name: "github.com",
url: "https://github.com/ollama/ollama",
expected: false,
},
{
name: "fake huggingface domain",
url: "https://nothuggingface.co/model",
expected: false,
},
{
name: "fake hf domain",
url: "https://nothf.co/model",
expected: false,
},
{
name: "huggingface in path not host",
url: "https://example.com/huggingface.co/model",
expected: false,
},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
var u *url.URL
if tc.url != "" {
var err error
u, err = url.Parse(tc.url)
if err != nil {
t.Fatalf("failed to parse URL: %v", err)
}
}
got := isHuggingFaceURL(u)
assert.Equal(t, tc.expected, got)
})
}
}
func TestGetNumDownloadParts(t *testing.T) {
tests := []struct {
name string
url string
envValue string
expected int
description string
}{
{
name: "nil url returns default",
url: "",
envValue: "",
expected: numDownloadParts,
description: "nil URL should return standard concurrency",
},
{
name: "ollama registry returns default",
url: "https://registry.ollama.ai/v2/library/llama3",
envValue: "",
expected: numDownloadParts,
description: "Ollama registry should use standard concurrency",
},
{
name: "huggingface returns reduced default",
url: "https://huggingface.co/model/repo",
envValue: "",
expected: numHFDownloadParts,
description: "HuggingFace should use reduced concurrency",
},
{
name: "hf.co CDN returns reduced default",
url: "https://cdn-lfs3.hf.co/repos/abc/123",
envValue: "",
expected: numHFDownloadParts,
description: "HuggingFace CDN should use reduced concurrency",
},
{
name: "huggingface with env override",
url: "https://huggingface.co/model/repo",
envValue: "2",
expected: 2,
description: "OLLAMA_HF_CONCURRENCY should override default",
},
{
name: "huggingface with higher env override",
url: "https://huggingface.co/model/repo",
envValue: "8",
expected: 8,
description: "OLLAMA_HF_CONCURRENCY can be set higher than default",
},
{
name: "huggingface with invalid env (non-numeric)",
url: "https://huggingface.co/model/repo",
envValue: "invalid",
expected: numHFDownloadParts,
description: "Invalid OLLAMA_HF_CONCURRENCY should fall back to default",
},
{
name: "huggingface with invalid env (zero)",
url: "https://huggingface.co/model/repo",
envValue: "0",
expected: numHFDownloadParts,
description: "Zero OLLAMA_HF_CONCURRENCY should fall back to default",
},
{
name: "huggingface with invalid env (negative)",
url: "https://huggingface.co/model/repo",
envValue: "-1",
expected: numHFDownloadParts,
description: "Negative OLLAMA_HF_CONCURRENCY should fall back to default",
},
{
name: "non-huggingface ignores env",
url: "https://registry.ollama.ai/v2/library/llama3",
envValue: "2",
expected: numDownloadParts,
description: "OLLAMA_HF_CONCURRENCY should not affect non-HF URLs",
},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
// Set or clear the environment variable
if tc.envValue != "" {
t.Setenv("OLLAMA_HF_CONCURRENCY", tc.envValue)
}
var u *url.URL
if tc.url != "" {
var err error
u, err = url.Parse(tc.url)
if err != nil {
t.Fatalf("failed to parse URL: %v", err)
}
}
got := getNumDownloadParts(u)
assert.Equal(t, tc.expected, got, tc.description)
})
}
}