repo stringlengths 5 53 | pr_number int32 1 321k | task_type stringclasses 2
values | issue_text stringlengths 0 81.2k | pr_title stringlengths 1 319 | pr_body stringlengths 0 105k | base_sha stringlengths 40 40 | head_sha stringlengths 40 40 | gold_diff stringlengths 0 202M | changed_files listlengths 0 100 | review_threads listlengths 0 100 | test_patch stringlengths 0 23.4M | merged bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
unslothai/unsloth | 6,267 | issue_to_patch | Studio: backfill the DiffusionGemma visual-server on a tag-matching update | ## Problem
On Mac (and in principle any platform) loading DiffusionGemma can fail with:
```
RuntimeError: DiffusionGemma runner not found...
```
even though the prebuilt bundle ships `llama-diffusion-gemma-visual-server`.
This happens when llama.cpp was installed before the visual-server was added to the install co... | 01d152b06f558ec4b63a62d4ea04b5ff28579ecc | 8629b5a1a7ea8ec38d8d7f75c2eb511d7e859cd6 | diff --git a/studio/install_llama_prebuilt.py b/studio/install_llama_prebuilt.py
index 219211034d..46aa14b041 100644
--- a/studio/install_llama_prebuilt.py
+++ b/studio/install_llama_prebuilt.py
@@ -6490,6 +6490,31 @@ def validate_prebuilt_attempts(
raise PrebuiltFallback("no prebuilt bundle passed validation")
... | [
"studio/install_llama_prebuilt.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Bypass the existing-install short circuit for backfills**\n\nIn the tag-matches-but-visual-server-is-missing case, this branch logs that it will re-extract and then falls through to `validate_prebuilt_attempts`... | true | ||
unslothai/unsloth | 6,264 | issue_to_patch | Studio: only advertise a Cloudflare tunnel once it actually serves | Follow-up to #6204 (auto Cloudflare quick tunnel for `0.0.0.0` launches). Three bugs let Studio print a `trycloudflare.com` link that doesn't work:
1. **Dead `api.` URL.** The URL regex `https://[A-Za-z0-9-]+\.trycloudflare\.com` also matches cloudflared's own API host in its failure line, so a transient API timeout... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 64c29977155c0581866a9180fb7b79e294311f63 | diff --git a/studio/backend/cloudflare_tunnel.py b/studio/backend/cloudflare_tunnel.py
index 7c088b6e1e..e5dba69452 100644
--- a/studio/backend/cloudflare_tunnel.py
+++ b/studio/backend/cloudflare_tunnel.py
@@ -24,12 +24,20 @@
from typing import Optional, Tuple
# cloudflared logs the quick-tunnel URL; match only th... | [
"studio/backend/cloudflare_tunnel.py",
"studio/backend/run.py",
"studio/backend/tests/test_cloudflare_tunnel.py"
] | [
{
"comment": "\n\nIf a concurrent shutdown (e.g., via `stop_studio_tunnel()`) or another concurrent start occurs while `wait_for_ready()` is blocking, `_active_tunnel` will be cleared or replaced. However, if `saw_url` is `True`, the loop will cu... | diff --git a/studio/backend/tests/test_cloudflare_tunnel.py b/studio/backend/tests/test_cloudflare_tunnel.py
index 873547631d..8042240b64 100644
--- a/studio/backend/tests/test_cloudflare_tunnel.py
+++ b/studio/backend/tests/test_cloudflare_tunnel.py
@@ -52,6 +52,26 @@ def test_url_regex_no_match_on_unrelated():
a... | true | |
unslothai/unsloth | 6,264 | comment_to_fix | Studio: only advertise a Cloudflare tunnel once it actually serves | 
If a concurrent shutdown (e.g., via `stop_studio_tunnel()`) or another concurrent start occurs while `wait_for_ready()` is blocking, `_active_tunnel` will be cleared or replaced. However, if `saw_url` is `True`, the loop will currently proceed to the n... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 64c29977155c0581866a9180fb7b79e294311f63 | diff --git a/studio/backend/cloudflare_tunnel.py b/studio/backend/cloudflare_tunnel.py
index 7c088b6e1e..e5dba69452 100644
--- a/studio/backend/cloudflare_tunnel.py
+++ b/studio/backend/cloudflare_tunnel.py
@@ -24,12 +24,20 @@
from typing import Optional, Tuple
# cloudflared logs the quick-tunnel URL; match only th... | [
"studio/backend/cloudflare_tunnel.py"
] | [
{
"comment": "\n\nIf a concurrent shutdown (e.g., via `stop_studio_tunnel()`) or another concurrent start occurs while `wait_for_ready()` is blocking, `_active_tunnel` will be cleared or replaced. However, if `saw_url` is `True`, the loop will cu... | true | ||
unslothai/unsloth | 6,264 | comment_to_fix | Studio: only advertise a Cloudflare tunnel once it actually serves | **<sub><sub></sub></sub> Track shutdown before retrying the tunnel**
When the first attempt mints a URL but times out, this retry loop clears `_active_tunnel` before continuing to the `http2` attempt. If `/api/shutdown` or another caller runs `stop_studio_... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 64c29977155c0581866a9180fb7b79e294311f63 | diff --git a/studio/backend/cloudflare_tunnel.py b/studio/backend/cloudflare_tunnel.py
index 7c088b6e1e..e5dba69452 100644
--- a/studio/backend/cloudflare_tunnel.py
+++ b/studio/backend/cloudflare_tunnel.py
@@ -24,12 +24,20 @@
from typing import Optional, Tuple
# cloudflared logs the quick-tunnel URL; match only th... | [
"studio/backend/cloudflare_tunnel.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Track shutdown before retrying the tunnel**\n\nWhen the first attempt mints a URL but times out, this retry loop clears `_active_tunnel` before continuing to the `http2` attempt. If `/api/shutdown` or another c... | true | ||
unslothai/unsloth | 6,266 | issue_to_patch | Studio: polish update banner layout and sidebar settings icon | ## Summary
Polishes the in-app update notification banners and the sidebar account menu.
### Update banners (web and llama.cpp)
- Add a download icon next to the title, aligned with the title text.
- Rework the action row so Release notes, Remind me later and Copy command are evenly spaced with balanced left and rig... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 9526ab32fd6e7fc0c03e21cd74f067741026ac1c | diff --git a/studio/frontend/src/app/provider.tsx b/studio/frontend/src/app/provider.tsx
index f806a0f405..39e04843a9 100644
--- a/studio/frontend/src/app/provider.tsx
+++ b/studio/frontend/src/app/provider.tsx
@@ -259,7 +259,7 @@ function TauriWrapper({ children }: { children: ReactNode }) {
<>
{childr... | [
"studio/frontend/src/app/provider.tsx",
"studio/frontend/src/components/app-sidebar.tsx",
"studio/frontend/src/components/llama-update-banner.tsx",
"studio/frontend/src/components/web/update-banner.tsx"
] | [
{
"comment": "**<sub><sub></sub></sub> Allow the update footer to wrap on mobile**\n\nOn narrow web viewports (for example 360–375px wide), this single nowrap flex row now contains the release-notes link plus two `Button`s, and the shared button sty... | true | ||
unslothai/unsloth | 5,841 | issue_to_patch | Studio: extend llama.cpp first-token timeout | ## Summary
Addresses #5756
Improves Studio behavior for slow local GGUF inference:
- Extend first-token / prefill timeout from 120s to 10 minutes.
- Remove Studio’s fixed `t_max_predict_ms` generation cutoff.
- Show a clearer first-token timeout message instead of generic lost connection.
- Add tests for th... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 75da00451e..80b9739cc1 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -122,15 +122,11 @@ def _wsl_system_rocm_lib_dirs() -> "list[str]":
)
_MAX_REPROM... | [
"studio/backend/core/inference/llama_cpp.py",
"studio/backend/routes/inference.py",
"studio/backend/tests/test_gguf_route_cursor_reset.py",
"studio/backend/tests/test_llama_cpp_tool_loop.py",
"studio/backend/tests/test_llama_route_timeouts.py",
"studio/backend/tests/test_openai_tool_passthrough.py",
"te... | [
{
"comment": "\n\nThe current implementation of `timeout_label` formatting does not handle singular/plural forms correctly if `_DEFAULT_FIRST_TOKEN_TIMEOUT_S` is changed to a value like `60.0` (which would output \"1 minutes\") or `1.0` (whic... | diff --git a/studio/backend/tests/test_gguf_route_cursor_reset.py b/studio/backend/tests/test_gguf_route_cursor_reset.py
index fdbf31f9de..dc397b2d08 100644
--- a/studio/backend/tests/test_gguf_route_cursor_reset.py
+++ b/studio/backend/tests/test_gguf_route_cursor_reset.py
@@ -59,11 +59,16 @@ def fake_stream_with_retr... | true | |
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | 
The current implementation of `timeout_label` formatting does not handle singular/plural forms correctly if `_DEFAULT_FIRST_TOKEN_TIMEOUT_S` is changed to a value like `60.0` (which would output "1 minutes") or `1.0` (which would output "1 seconds"... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "\n\nThe current implementation of `timeout_label` formatting does not handle singular/plural forms correctly if `_DEFAULT_FIRST_TOKEN_TIMEOUT_S` is changed to a value like `60.0` (which would output \"1 minutes\") or `1.0` (whic... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | Concern: this branch routes every `httpx.ReadTimeout` to a hard-coded "did not produce a first token within ..." message, but `ReadTimeout` can also fire mid-stream, not only during prefill. In `_openai_passthrough_stream` (around line 5943) a stall during `aiter_lines()` after some tokens have already been emitted lan... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "Concern: this branch routes every `httpx.ReadTimeout` to a hard-coded \"did not produce a first token within ...\" message, but `ReadTimeout` can also fire mid-stream, not only during prefill. In `_openai_passthrough_stream` (around line 5943) a stall during `aiter_lines()` after some tokens have ... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | Heads up: combining this 600s read timeout with the removal of `payload["t_max_predict_ms"]` at the three payload-builder sites in this PR creates an unbounded generation hang.
`_iter_text_cancellable` at lines 4260-4285 catches `httpx.ReadTimeout` and `continue`s indefinitely. Its docstring (and the surrounding code)... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 75da00451e..80b9739cc1 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -122,15 +122,11 @@ def _wsl_system_rocm_lib_dirs() -> "list[str]":
)
_MAX_REPROM... | [
"studio/backend/core/inference/llama_cpp.py"
] | [
{
"comment": "Heads up: combining this 600s read timeout with the removal of `payload[\"t_max_predict_ms\"]` at the three payload-builder sites in this PR creates an unbounded generation hang.\n\n`_iter_text_cancellable` at lines 4260-4285 catches `httpx.ReadTimeout` and `continue`s indefinitely. Its docstring ... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Keep Stop responsive during long prefill**
When a user hits Stop while llama-server is still in prefill and has not returned response headers, `_cancel_watcher` has no response object to close, so the main thread remains blocked in... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 75da00451e..80b9739cc1 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -122,15 +122,11 @@ def _wsl_system_rocm_lib_dirs() -> "list[str]":
)
_MAX_REPROM... | [
"studio/backend/core/inference/llama_cpp.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Keep Stop responsive during long prefill**\n\nWhen a user hits Stop while llama-server is still in prefill and has not returned response headers, `_cancel_watcher` has no response object to close, so the main t... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Avoid timing out long non-streaming generations**
For non-streaming OpenAI chat passthrough requests (the tools/response_format path where `payload.stream` is false), this 600s read timeout is now the only wall-clock limit after th... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Avoid timing out long non-streaming generations**\n\nFor non-streaming OpenAI chat passthrough requests (the tools/response_format path where `payload.stream` is false), this 600s read timeout is now the only w... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Make cancel observe pre-header passthrough sends**
In the streaming tools/response_format passthrough path, the route awaits `client.send(..., stream=True)` before returning the `StreamingResponse` and before `_stream()` starts `_a... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Make cancel observe pre-header passthrough sends**\n\nIn the streaming tools/response_format passthrough path, the route awaits `client.send(..., stream=True)` before returning the `StreamingResponse` and befor... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Keep streamed body reads on a stall timeout**
Passing the 20-minute first-token timeout as the scalar `AsyncClient` timeout also makes every subsequent streamed body read wait up to 20 minutes. In the `/v1/completions` streaming pr... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Keep streamed body reads on a stall timeout**\n\nPassing the 20-minute first-token timeout as the scalar `AsyncClient` timeout also makes every subsequent streamed body read wait up to 20 minutes. In the `/v1/c... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Cap raw completions when removing read timeout**
For non-streaming `/v1/completions` calls that omit `max_tokens`, this route forwards the raw request body unchanged and now uses an uncapped response-body read. llama-server default... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Cap raw completions when removing read timeout**\n\nFor non-streaming `/v1/completions` calls that omit `max_tokens`, this route forwards the raw request body unchanged and now uses an uncapped response-body re... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Race streamed completions against disconnects**
When `/v1/completions` is streamed and llama-server is still in prefill before response headers, the only disconnect check is inside `_aiter_llama_stream_items`, which is not reached ... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Race streamed completions against disconnects**\n\nWhen `/v1/completions` is streamed and llama-server is still in prefill before response headers, the only disconnect check is inside `_aiter_llama_stream_items... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Stop Responses streams before headers arrive**
For streamed `/v1/responses`, the preamble is yielded before the upstream send, but if the client disconnects during a long llama-server prefill the generator remains stuck awaiting `c... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Stop Responses streams before headers arrive**\n\nFor streamed `/v1/responses`, the preamble is yielded before the upstream send, but if the client disconnects during a long llama-server prefill the generator r... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Cancel child send tasks on request cancellation**
When the request task itself is cancelled while this helper is waiting for headers, the helper's `finally` only cancels `cancel_task`; the background `send_task` created here is lef... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Cancel child send tasks on request cancellation**\n\nWhen the request task itself is cancelled while this helper is waiting for headers, the helper's `finally` only cancels `cancel_task`; the background `send_t... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Use Python 3.9-compatible async iteration**
In Python 3.9 deployments, which are still allowed by this repo's `requires-python = ">=3.9,<3.15"`, `anext` is not a builtin, so any streamed llama-server path that reaches this helper r... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Use Python 3.9-compatible async iteration**\n\nIn Python 3.9 deployments, which are still allowed by this repo's `requires-python = \">=3.9,<3.15\"`, `anext` is not a builtin, so any streamed llama-server path ... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Keep a read deadline for non-streaming requests**
For non-streaming `/v1/completions` and the chat/Anthropic passthrough paths that use this helper, a llama-server request that accepts the connection but then spends forever in pref... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Keep a read deadline for non-streaming requests**\n\nFor non-streaming `/v1/completions` and the chat/Anthropic passthrough paths that use this helper, a llama-server request that accepts the connection but the... | true | ||
unslothai/unsloth | 5,841 | comment_to_fix | Studio: extend llama.cpp first-token timeout | **<sub><sub></sub></sub> Enforce the first-token deadline after headers**
When an upstream streaming response sends HTTP headers before the first SSE/body item, this await can still block for the full 20-minute httpx read timeout even if the `first_token_d... | b91116cacc1c6fe18fb0c2455ef6bbdecaf08ab1 | 01603b888b5e6f08b99634304f973f29e6a4894c | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fa708d0e39..e9353cd803 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -121,6 +121,19 @@ def _template_raise_message(error_text: str, chat_template: Optional[str]) -> Op
def _friendl... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Enforce the first-token deadline after headers**\n\nWhen an upstream streaming response sends HTTP headers before the first SSE/body item, this await can still block for the full 20-minute httpx read timeout ev... | true | ||
unslothai/unsloth | 6,262 | issue_to_patch | studio: declare UNSLOTH_IS_PRESENT at backend startup (clean-install + Windows) | ## Summary
Follow-up to #6259. That PR set `UNSLOTH_IS_PRESENT` for the DiffusionGemma shim subprocess, but the studio backend process has the same problem: it lazily imports `unsloth_zoo` submodules in several places, and each import of the top-level `unsloth_zoo` raises
```
ImportError: Please install Unsloth via `... | cc32777f30df414d0457aaf7e321e92a2a5a6908 | 02482b952cdbea1f32773508ac313986acc99cc5 | diff --git a/studio/backend/main.py b/studio/backend/main.py
index 2b34b11fef..064e261753 100644
--- a/studio/backend/main.py
+++ b/studio/backend/main.py
@@ -193,6 +193,11 @@ def _ver_key(name: str) -> tuple:
if not os.environ.get("UNSLOTH_LLAMA_CPP_PATH"):
os.environ["UNSLOTH_LLAMA_CPP_PATH"] = str(_STU... | [
"studio/backend/main.py",
"studio/backend/run.py"
] | [] | true | ||
unslothai/unsloth | 5,729 | issue_to_patch | [Bug] Studio does not properly list ByteSape quants
There are three IQ4_XS quants, but Unsloth Studio merged them (and their size):
<img width="403" height="160" alt="Image" src="https://github.com/user-attachments/assets/58f95303-e534-44e8-859c-697458f63a5a" />
Studio Version v0.1.415-beta
Package Version 2026.5.6 | Studio: keep distinct bpw flavors of the same GGUF quant | Fixes #5728.
### Problem
Studio's quant picker merges files that share a base quant label but differ in bits-per-weight. Repro repo from the issue: `byteshape/Qwen3.6-35B-A3B-MTP-GGUF` ships three IQ4_XS files (3.53 / 3.97 / 4.19 bpw), and Studio shows a single `IQ4_XS` row with their summed size (~48 GB).
###... | 5300c047b65edb8dfb2712541c88a1540fb879da | 5ec17a71fee2947a605035406bc0704994be8093 | diff --git a/studio/backend/utils/models/model_config.py b/studio/backend/utils/models/model_config.py
index 3381ae0391..f91b8fb8c9 100644
--- a/studio/backend/utils/models/model_config.py
+++ b/studio/backend/utils/models/model_config.py
@@ -1343,6 +1343,7 @@ def _extract_quant_label(filename: str) -> str:
"m... | [
"studio/backend/utils/models/model_config.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Update the active GGUF variant parser too**\n\nThis adds bpw parsing only to `utils/models/model_config.py`, but the Studio quant picker is served by `/api/hub/gguf-variants` (`studio/backend/hub/routes/invento... | true | |
unslothai/unsloth | 6,225 | issue_to_patch | Installer: drop the lemonade ROCm fallback now the fork ships identical per-gfx prebuilts | #5963 made `unslothai/llama.cpp` ship per-gfx ROCm bundles from a workflow that's a literal copy of lemonade-sdk's, so the binaries and gfx coverage are identical. The `lemonade-sdk/llamacpp-rocm` fallback we kept alongside them is now redundant, and it cost an extra `api.github.com` call on every ROCm install (the lem... | 3427e3fd6236873d07cfa334778b6384541e8868 | 3c7e344097a0762a282b0a464f0a1fa0a5420ba3 | diff --git a/install.ps1 b/install.ps1
index 9abddc9ce2..562e44231c 100644
--- a/install.ps1
+++ b/install.ps1
@@ -1574,7 +1574,7 @@ shell.Run cmd, 0, False
# popping a UAC/DiskPart prompt RunAsInvoker can't suppress (manifest is
# asInvoker). So only probe when a HIP SDK is present (hipinfo found ->
... | [
"install.ps1",
"studio/backend/tests/test_lemonade_llamacpp_rocm_bins_mock.py",
"studio/backend/tests/test_llama_cpp_update.py",
"studio/backend/utils/llama_cpp_update.py",
"studio/install_llama_prebuilt.py",
"studio/install_python_stack.py",
"studio/setup.ps1",
"studio/setup.sh",
"tests/studio/inst... | [
{
"comment": "**<sub><sub></sub></sub> Preserve ROCm family targets during updates**\n\nWhen updating an existing per-gfx ROCm install, the backend derives `--rocm-gfx gfx110x`/`gfx103x` from marker assets like `app-...-rocm-gfx110X.tar.gz`; after t... | diff --git a/studio/backend/tests/test_lemonade_llamacpp_rocm_bins_mock.py b/studio/backend/tests/test_lemonade_llamacpp_rocm_bins_mock.py
deleted file mode 100644
index 7c6c514b8f..0000000000
--- a/studio/backend/tests/test_lemonade_llamacpp_rocm_bins_mock.py
+++ /dev/null
@@ -1,523 +0,0 @@
-# SPDX-License-Identifier:... | true | |
unslothai/unsloth | 6,225 | comment_to_fix | Installer: drop the lemonade ROCm fallback now the fork ships identical per-gfx prebuilts | **<sub><sub></sub></sub> Preserve ROCm family targets during updates**
When updating an existing per-gfx ROCm install, the backend derives `--rocm-gfx gfx110x`/`gfx103x` from marker assets like `app-...-rocm-gfx110X.tar.gz`; after this fallback removal, th... | 3427e3fd6236873d07cfa334778b6384541e8868 | 3c7e344097a0762a282b0a464f0a1fa0a5420ba3 | diff --git a/studio/install_llama_prebuilt.py b/studio/install_llama_prebuilt.py
index 680b621b18..d98bb339a3 100644
--- a/studio/install_llama_prebuilt.py
+++ b/studio/install_llama_prebuilt.py
@@ -9,7 +9,6 @@
import argparse
import errno
import fnmatch
-import functools
import hashlib
import json
import os
@@ -... | [
"studio/install_llama_prebuilt.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Preserve ROCm family targets during updates**\n\nWhen updating an existing per-gfx ROCm install, the backend derives `--rocm-gfx gfx110x`/`gfx103x` from marker assets like `app-...-rocm-gfx110X.tar.gz`; after t... | true | ||
unslothai/unsloth | 6,225 | comment_to_fix | Installer: drop the lemonade ROCm fallback now the fork ships identical per-gfx prebuilts | **<sub><sub></sub></sub> Route inferred ROCm GPUs to the fork**
When Windows only has an inferred `$script:ROCmGfxArch` (the Adrenalin-only/name-inferred case described here) and `$HasROCm` is still false, `$HelperReleaseRepo` has already been set to `ggml... | 3427e3fd6236873d07cfa334778b6384541e8868 | 3c7e344097a0762a282b0a464f0a1fa0a5420ba3 | diff --git a/studio/setup.ps1 b/studio/setup.ps1
index 94817b561a..75adfa5b15 100644
--- a/studio/setup.ps1
+++ b/studio/setup.ps1
@@ -850,7 +850,7 @@ if (-not $HasNvidiaSmi) {
# popping a UAC/DiskPart prompt RunAsInvoker can't suppress (its manifest is
# asInvoker; even 'amd-smi version' hangs). So only prob... | [
"studio/setup.ps1"
] | [
{
"comment": "**<sub><sub></sub></sub> Route inferred ROCm GPUs to the fork**\n\nWhen Windows only has an inferred `$script:ROCmGfxArch` (the Adrenalin-only/name-inferred case described here) and `$HasROCm` is still false, `$HelperReleaseRepo` has a... | true | ||
unslothai/unsloth | 6,225 | comment_to_fix | Installer: drop the lemonade ROCm fallback now the fork ships identical per-gfx prebuilts | **<sub><sub></sub></sub> Honor env-provided ROCm gfx when routing**
On Linux hosts with no ROCm tooling on `PATH` where the user supplies `UNSLOTH_ROCM_GFX_ARCH`, `_setup_gfx` is never populated because the env override is only read inside the AMD-detected... | 3427e3fd6236873d07cfa334778b6384541e8868 | 3c7e344097a0762a282b0a464f0a1fa0a5420ba3 | diff --git a/studio/setup.sh b/studio/setup.sh
index a8603bd0da..ea98bf441e 100755
--- a/studio/setup.sh
+++ b/studio/setup.sh
@@ -340,9 +340,9 @@ binary_tag = str(payload.get("binary_release_tag") or "").strip()
if not repo or not release_tag:
raise SystemExit(0)
-# For non-upstream sources (e.g. lemonade) the... | [
"studio/setup.sh"
] | [
{
"comment": "**<sub><sub></sub></sub> Honor env-provided ROCm gfx when routing**\n\nOn Linux hosts with no ROCm tooling on `PATH` where the user supplies `UNSLOTH_ROCM_GFX_ARCH`, `_setup_gfx` is never populated because the env override is only read... | true | ||
unslothai/unsloth | 6,260 | issue_to_patch | Bump install.sh / install.ps1 pin to unsloth>=2026.6.5 | ## Summary
- PyPI release [unsloth 2026.6.5](https://pypi.org/project/unsloth/2026.6.5/) is now live.
- Bumps the pinned floor in `install.sh` and `install.ps1` from `unsloth>=2026.6.4` to `unsloth>=2026.6.5` so fresh installs resolve to the new wheel.
- Tagged on main as `v0.1.462-beta`.
## Test plan
- [ ] Sanity-r... | c2b8da704d57983550fcb0b6eade5e2d3965b4d1 | 7a9f7f3230986146a60a4c04947ced96556a047e | diff --git a/install.ps1 b/install.ps1
index 515f700fea..189857d8b4 100644
--- a/install.ps1
+++ b/install.ps1
@@ -1921,7 +1921,7 @@ shell.Run cmd, 0, False
if ($SkipTorch) {
# No-torch: install unsloth + unsloth-zoo with --no-deps, then
# runtime deps (typer, safetensors, transformer... | [
"install.ps1",
"install.sh"
] | [] | true | ||
unslothai/unsloth | 5,628 | issue_to_patch | studio/frontend: model picker list lacks arrow-key keyboard navigation
## Repro
1. Open Studio /chat with no mouse plugged in.
2. Click anywhere in the page (or Tab) and Tab to the model picker button in the chat header.
3. Press **Enter** -- the picker dialog opens with focus on the "Hub models" tab.
4. Press **Arro... | studio: add keyboard navigation to model picker | ## What changed
- Added listbox roles for the Hub and Fine-tuned model lists.
- Added roving tab index, so Tab goes into one model row.
- ArrowUp, ArrowDown, Home, and End now moves focus in the list.
- Expanded GGUF variants can be reached with ArrowDown from the parent row.
- Variant rows and visible delete buttons c... | be2a122e210c8e4033e18e52d9d0cbb26f7d9310 | c4eab944ff3a3e7bff326c37bcb51f187f432f3e | diff --git a/studio/frontend/src/components/assistant-ui/model-selector.tsx b/studio/frontend/src/components/assistant-ui/model-selector.tsx
index fa6980824f..b97c34152f 100644
--- a/studio/frontend/src/components/assistant-ui/model-selector.tsx
+++ b/studio/frontend/src/components/assistant-ui/model-selector.tsx
@@ -2... | [
"studio/frontend/src/components/assistant-ui/model-selector.tsx",
"studio/frontend/src/components/assistant-ui/model-selector/pickers.tsx",
"studio/frontend/src/features/hub/lib/gguf-fit.ts"
] | [
{
"comment": "\n\nThe `useRovingModelList` hook provides a solid foundation for keyboard navigation, but it currently only handles a flat list of options. In the `HubModelPicker` and `LoraModelPicker`, some items (GGUF models) can be expanded... | true | |
unslothai/unsloth | 5,628 | comment_to_fix | studio: add keyboard navigation to model picker | 
The `useRovingModelList` hook provides a solid foundation for keyboard navigation, but it currently only handles a flat list of options. In the `HubModelPicker` and `LoraModelPicker`, some items (GGUF models) can be expanded to show variants, and s... | be2a122e210c8e4033e18e52d9d0cbb26f7d9310 | c4eab944ff3a3e7bff326c37bcb51f187f432f3e | diff --git a/studio/frontend/src/components/assistant-ui/model-selector/pickers.tsx b/studio/frontend/src/components/assistant-ui/model-selector/pickers.tsx
index 0afabb6423..3165057021 100644
--- a/studio/frontend/src/components/assistant-ui/model-selector/pickers.tsx
+++ b/studio/frontend/src/components/assistant-ui/... | [
"studio/frontend/src/components/assistant-ui/model-selector/pickers.tsx"
] | [
{
"comment": "\n\nThe `useRovingModelList` hook provides a solid foundation for keyboard navigation, but it currently only handles a flat list of options. In the `HubModelPicker` and `LoraModelPicker`, some items (GGUF models) can be expanded... | true | ||
unslothai/unsloth | 5,628 | comment_to_fix | studio: add keyboard navigation to model picker | 
The `handlePickerEntryKeyDown` function currently prevents `ArrowDown` navigation if the focus is inside an `input`. While this prevents interference with standard input behavior, it also prevents users from using `ArrowDown` to jump from the searc... | be2a122e210c8e4033e18e52d9d0cbb26f7d9310 | c4eab944ff3a3e7bff326c37bcb51f187f432f3e | diff --git a/studio/frontend/src/components/assistant-ui/model-selector.tsx b/studio/frontend/src/components/assistant-ui/model-selector.tsx
index fa6980824f..b97c34152f 100644
--- a/studio/frontend/src/components/assistant-ui/model-selector.tsx
+++ b/studio/frontend/src/components/assistant-ui/model-selector.tsx
@@ -2... | [
"studio/frontend/src/components/assistant-ui/model-selector.tsx"
] | [
{
"comment": "\n\nThe `handlePickerEntryKeyDown` function currently prevents `ArrowDown` navigation if the focus is inside an `input`. While this prevents interference with standard input behavior, it also prevents users from using `ArrowDown... | true | ||
unslothai/unsloth | 6,259 | issue_to_patch | DiffusionGemma: set UNSLOTH_IS_PRESENT so the shim runs on a clean install | ## Summary
A clean install cannot run DiffusionGemma. The runner spawns `python -m unsloth_zoo.diffusion_studio.shim`, and `unsloth_zoo/__init__.py` refuses to import unless `UNSLOTH_IS_PRESENT` is in the environment (normally set by `import unsloth`). The shim never imports `unsloth`, so on a fresh PyPI/install.sh se... | 4c4422893cab37225c4ea422f6d643e6cef14087 | 18f0ba19e180fc93b3932e3c7a2cc601447873ee | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 19d02d36fe..75da00451e 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -2527,6 +2527,10 @@ def _start_diffusion_server(
]
env = child... | [
"studio/backend/core/inference/llama_cpp.py"
] | [] | true | ||
unslothai/unsloth | 6,255 | issue_to_patch | DiffusionGemma Studio: disable tools, enable artifacts canvas by default | ## Summary
Three fixes so DiffusionGemma works end to end in Studio from a clean release install.
## 1. Set UNSLOTH_IS_PRESENT for the shim subprocess (fresh-install blocker)
The runner spawns `python -m unsloth_zoo.diffusion_studio.shim`. `unsloth_zoo/__init__.py` refuses to import unless `UNSLOTH_IS_PRESENT` is in... | 587e59ca83f5ef301a571665bc66bc24746fd119 | 361854189a7530b673bfb174f81be79a09c0a899 | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index b5d5328fb0..f30a8acf89 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -1042,6 +1042,10 @@ def _request_reasoning_kwargs(
@property
def suppo... | [
"studio/backend/core/inference/llama_cpp.py",
"studio/frontend/src/components/assistant-ui/markdown-text.tsx"
] | [] | true | ||
unslothai/unsloth | 6,258 | issue_to_patch | install.sh: keep the studio launch from draining the curl | sh script (WSL/dash) | ## Summary
Installing with `curl -fsSL https://unsloth.ai/install.sh | sh` can end with:
```
sh: 2913: Syntax error: end of file unexpected (expecting "fi")
```
The install itself completes (Studio installs, the prebuilt attaches), but the script dies right at the end. Reported on WSL, where `/bin/sh` is `dash`.
##... | 4c4422893cab37225c4ea422f6d643e6cef14087 | 46336cc08952d52546546a6f2d415bf6ab066d76 | diff --git a/install.sh b/install.sh
index af062ae057..d2432fb750 100755
--- a/install.sh
+++ b/install.sh
@@ -2900,7 +2900,10 @@ if [ -t 1 ]; then
case "${_reply:-y}" in
[Yy]*|"")
step "launch" "starting Unsloth Studio..."
- "$VENV_DIR/bin/unsloth" studio -p 8888
+ # De... | [
"install.sh"
] | [] | true | ||
unslothai/unsloth | 5,770 | issue_to_patch | Studio treats local GGUF+mmproj as a Transformers vision model and fails looking for config.json
### Summary
Unsloth Studio can successfully load a local GGUF Qwen3.6 model through `llama-server`, including `mmproj` and MTP speculative decoding, but then also routes the same `.gguf` file through the Transformers/Unsl... | fix(studio): keep local GGUF vision on llama-server | Fixes #5768
Summary
- use the existing mmproj detection for local GGUF vision capability checks, instead of passing `.gguf` files into Transformers config detection
- keep the older UI-selection path on the seperate llama-server GGUF route
- cover Qwen3.6 GGUFs with and without mmproj, including a projector added afte... | 3427e3fd6236873d07cfa334778b6384541e8868 | 17ea898156fca7974358d5b55b31f3a67791901b | diff --git a/studio/backend/utils/models/model_config.py b/studio/backend/utils/models/model_config.py
index d87fb6aa09..3381ae0391 100644
--- a/studio/backend/utils/models/model_config.py
+++ b/studio/backend/utils/models/model_config.py
@@ -721,6 +721,25 @@ def is_vision_model(model_name: str, hf_token: Optional[str]... | [
"studio/backend/tests/test_vision_cache.py",
"studio/backend/utils/models/model_config.py"
] | [
{
"comment": "\n\nSince local GGUF vision checks are explicitly uncached to allow for projectors being added after the initial inspection, this `logger.info` will fire on every call to `is_vision_model` for local GGUF models. This can lead to... | diff --git a/studio/backend/tests/test_vision_cache.py b/studio/backend/tests/test_vision_cache.py
index 2fee50d842..d1bdec8449 100644
--- a/studio/backend/tests/test_vision_cache.py
+++ b/studio/backend/tests/test_vision_cache.py
@@ -30,6 +30,7 @@
sys.modules.setdefault("loggers", _loggers_stub)
from utils.models.... | true |
unslothai/unsloth | 5,770 | comment_to_fix | fix(studio): keep local GGUF vision on llama-server | 
Since local GGUF vision checks are explicitly uncached to allow for projectors being added after the initial inspection, this `logger.info` will fire on every call to `is_vision_model` for local GGUF models. This can lead to excessive log noise if ... | 3427e3fd6236873d07cfa334778b6384541e8868 | 17ea898156fca7974358d5b55b31f3a67791901b | diff --git a/studio/backend/utils/models/model_config.py b/studio/backend/utils/models/model_config.py
index d87fb6aa09..3381ae0391 100644
--- a/studio/backend/utils/models/model_config.py
+++ b/studio/backend/utils/models/model_config.py
@@ -721,6 +721,25 @@ def is_vision_model(model_name: str, hf_token: Optional[str]... | [
"studio/backend/utils/models/model_config.py"
] | [
{
"comment": "\n\nSince local GGUF vision checks are explicitly uncached to allow for projectors being added after the initial inspection, this `logger.info` will fire on every call to `is_vision_model` for local GGUF models. This can lead to... | true | ||
unslothai/unsloth | 5,770 | comment_to_fix | fix(studio): keep local GGUF vision on llama-server | **<sub><sub></sub></sub> Search the GGUF snapshot root for direct-file selections**
When `model_name` is a direct GGUF file inside a variant subdirectory (for example a cached/exported layout like `snapshot/BF16/model.gguf` with `mmproj-*.gguf` stored at `... | 3427e3fd6236873d07cfa334778b6384541e8868 | 17ea898156fca7974358d5b55b31f3a67791901b | diff --git a/studio/backend/utils/models/model_config.py b/studio/backend/utils/models/model_config.py
index d87fb6aa09..3381ae0391 100644
--- a/studio/backend/utils/models/model_config.py
+++ b/studio/backend/utils/models/model_config.py
@@ -721,6 +721,25 @@ def is_vision_model(model_name: str, hf_token: Optional[str]... | [
"studio/backend/utils/models/model_config.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Search the GGUF snapshot root for direct-file selections**\n\nWhen `model_name` is a direct GGUF file inside a variant subdirectory (for example a cached/exported layout like `snapshot/BF16/model.gguf` with `mm... | true | ||
unslothai/unsloth | 5,849 | issue_to_patch | Studio: account for mmproj VRAM in GGUF fit budget (#5825) | ## Problem
Vision GGUFs load the mmproj (vision projector) onto the GPU via `--mmproj` alongside the weights, but Studio's context auto-sizing / GPU-selection budget in `LlamaCppBackend.load_model` sizes off `_get_gguf_size_bytes(model_path)`, which counts only the weight file(s). The projector is never added, so the ... | 7397c8784390c3f19b114bf64204db2714aa2a88 | 507b69433730349958e353a2bded50c3ebff036b | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index f30a8acf89..19d02d36fe 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -2996,6 +2996,16 @@ def _resolve_launch_mmproj_path(
return str(mmproj)... | [
"studio/backend/core/inference/llama_cpp.py",
"studio/backend/tests/test_mmproj_vram_accounting.py"
] | [
{
"comment": "\n\nMixing explicit string concatenation (`+`) with implicit string literal concatenation inside parentheses is hard to read and prone to formatting/spacing errors. We can simplify this by using a nested f-string expression dire... | diff --git a/studio/backend/tests/test_mmproj_vram_accounting.py b/studio/backend/tests/test_mmproj_vram_accounting.py
new file mode 100644
index 0000000000..bee289f183
--- /dev/null
+++ b/studio/backend/tests/test_mmproj_vram_accounting.py
@@ -0,0 +1,40 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-pr... | true | |
unslothai/unsloth | 5,849 | comment_to_fix | Studio: account for mmproj VRAM in GGUF fit budget (#5825) | 
Mixing explicit string concatenation (`+`) with implicit string literal concatenation inside parentheses is hard to read and prone to formatting/spacing errors. We can simplify this by using a nested f-string expression directly within the implicit... | 7397c8784390c3f19b114bf64204db2714aa2a88 | 507b69433730349958e353a2bded50c3ebff036b | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index f30a8acf89..19d02d36fe 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -2996,6 +2996,16 @@ def _resolve_launch_mmproj_path(
return str(mmproj)... | [
"studio/backend/core/inference/llama_cpp.py"
] | [
{
"comment": "\n\nMixing explicit string concatenation (`+`) with implicit string literal concatenation inside parentheses is hard to read and prone to formatting/spacing errors. We can simplify this by using a nested f-string expression dire... | true | ||
unslothai/unsloth | 6,257 | issue_to_patch | Bump install.sh / install.ps1 pin to unsloth>=2026.6.4 | ## Summary
- PyPI release [unsloth 2026.6.4](https://pypi.org/project/unsloth/2026.6.4/) is now live.
- Bumps the pinned floor in `install.sh` and `install.ps1` from `unsloth>=2026.6.3` to `unsloth>=2026.6.4` so fresh installs resolve to the new wheel.
- Tagged on main as `v0.1.46-beta`.
## Test plan
- [ ] Sanity-ru... | 756265e3fd57ea8b5516eb361e44162ca971affc | 554abe446dcc24e21ba63c303d4bce48729e7fc6 | diff --git a/install.ps1 b/install.ps1
index 9abddc9ce2..515f700fea 100644
--- a/install.ps1
+++ b/install.ps1
@@ -1921,7 +1921,7 @@ shell.Run cmd, 0, False
if ($SkipTorch) {
# No-torch: install unsloth + unsloth-zoo with --no-deps, then
# runtime deps (typer, safetensors, transformer... | [
"install.ps1",
"install.sh"
] | [] | true | ||
unslothai/unsloth | 6,253 | issue_to_patch | Studio: clearer check mark and even plus-menu hover padding | Two small Studio menu polish tweaks.
## Clearer app-wide check mark
Every check mark now renders through a shared `tick-icon` module instead of importing the icon directly in each component. The geometry is a plain tick sized a touch larger than the stock icon so it reads clearly in menus (plus menu toggles, checkboxe... | ce34aba932d9408fb32877e8f2deeb69933edca7 | db9942a62bb5690f2840237c60877c948f443fb5 | diff --git a/studio/frontend/src/components/assistant-ui/markdown-text.tsx b/studio/frontend/src/components/assistant-ui/markdown-text.tsx
index 45a0082a57..02418753fa 100644
--- a/studio/frontend/src/components/assistant-ui/markdown-text.tsx
+++ b/studio/frontend/src/components/assistant-ui/markdown-text.tsx
@@ -8,7 +... | [
"studio/frontend/src/components/assistant-ui/markdown-text.tsx",
"studio/frontend/src/components/assistant-ui/thread.tsx",
"studio/frontend/src/components/ui/checkbox.tsx",
"studio/frontend/src/components/ui/combobox.tsx",
"studio/frontend/src/components/ui/command.tsx",
"studio/frontend/src/components/ui... | [
{
"comment": "\n\nHardcoding `strokeWidth: \"2\"` on the path attributes overrides any `strokeWidth` prop passed to the `HugeiconsIcon` component (such as `strokeWidth={1.75}` in `markdown-text.tsx`, `thread.tsx`, and `path-info-button.tsx`).... | true | ||
unslothai/unsloth | 6,205 | issue_to_patch | Studio: project sources backed by RAG | ## What this does
The Sources tab on projects was a placeholder with a disabled Add sources button. This PR makes project sources fully functional on top of the RAG engine, and cleans up the project UI.
Uploading documents to a project indexes them for retrieval, and every chat inside that project can then reference ... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/backend/core/inference/tools.py b/studio/backend/core/inference/tools.py
index 8c45706395..43c9610282 100644
--- a/studio/backend/core/inference/tools.py
+++ b/studio/backend/core/inference/tools.py
@@ -806,6 +806,7 @@ def _search_knowledge_base(arguments: dict, rag_scope: dict | None) -> str:
... | [
"studio/backend/core/inference/tools.py",
"studio/backend/core/rag/ingestion.py",
"studio/backend/core/rag/retrieval.py",
"studio/backend/core/rag/store.py",
"studio/backend/core/rag/tool.py",
"studio/backend/routes/chat_history.py",
"studio/backend/routes/rag.py",
"studio/backend/storage/rag_db.py",
... | [
{
"comment": "\n\nWhen deleting a project, its associated RAG documents are deleted from the database. However, the physical files stored on disk at `stored_path` are left orphaned, leading to a disk space leak over time. We should also delete th... | diff --git a/studio/backend/tests/test_rag_ingestion.py b/studio/backend/tests/test_rag_ingestion.py
index c5fe15876d..f0b71bc23b 100644
--- a/studio/backend/tests/test_rag_ingestion.py
+++ b/studio/backend/tests/test_rag_ingestion.py
@@ -39,7 +39,7 @@ def test_ingestion_lifecycle_pending_to_completed(rag_home, stub_em... | true | |
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | 
When deleting a project, its associated RAG documents are deleted from the database. However, the physical files stored on disk at `stored_path` are left orphaned, leading to a disk space leak over time. We should also delete the physical files from di... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/backend/routes/chat_history.py b/studio/backend/routes/chat_history.py
index 782bba3be0..2ea572e30e 100644
--- a/studio/backend/routes/chat_history.py
+++ b/studio/backend/routes/chat_history.py
@@ -332,6 +332,35 @@ async def delete_project(
status_code = 404,
detail = f"Pr... | [
"studio/backend/routes/chat_history.py"
] | [
{
"comment": "\n\nWhen deleting a project, its associated RAG documents are deleted from the database. However, the physical files stored on disk at `stored_path` are left orphaned, leading to a disk space leak over time. We should also delete th... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | 
If `scope` is empty (e.g., an empty list or iterable), `_scopes(scope)` will return an empty list. This causes `placeholders` to be an empty string, resulting in a SQL syntax error `scope IN ()` in SQLite. We should defensively guard against an emp... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/backend/core/rag/store.py b/studio/backend/core/rag/store.py
index a77b9af7f8..7d58931e53 100644
--- a/studio/backend/core/rag/store.py
+++ b/studio/backend/core/rag/store.py
@@ -29,6 +29,15 @@ def thread_scope(thread_id: str) -> str:
return f"thread_{thread_id}"
+def project_scope(project_... | [
"studio/backend/core/rag/store.py"
] | [
{
"comment": "\n\nIf `scope` is empty (e.g., an empty list or iterable), `_scopes(scope)` will return an empty list. This causes `placeholders` to be an empty string, resulting in a SQL syntax error `scope IN ()` in SQLite. We should defensiv... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | 
The project card `div` is interactive and clickable, but it lacks accessibility attributes. It should have `role="button"`, `tabIndex={0}`, and an `onKeyDown` handler to support keyboard navigation (Enter and Space keys) for users with assistive te... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/frontend/src/features/chat/projects-page.tsx b/studio/frontend/src/features/chat/projects-page.tsx
index 724cd0a755..87fe9b66ec 100644
--- a/studio/frontend/src/features/chat/projects-page.tsx
+++ b/studio/frontend/src/features/chat/projects-page.tsx
@@ -41,6 +41,7 @@ import {
Delete02Icon,
Do... | [
"studio/frontend/src/features/chat/projects-page.tsx"
] | [
{
"comment": "\n\nThe project card `div` is interactive and clickable, but it lacks accessibility attributes. It should have `role=\"button\"`, `tabIndex={0}`, and an `onKeyDown` handler to support keyboard navigation (Enter and Space keys) f... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | **<sub><sub></sub></sub> Refresh source cache after uploads settle**
When a user adds project sources and submits a chat before `upload(files)` has created the document row, `projectHasSources()` can observe an empty list and cache `false` for 30 seconds; ... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/frontend/src/features/rag/components/project-sources-panel.tsx b/studio/frontend/src/features/rag/components/project-sources-panel.tsx
new file mode 100644
index 0000000000..941e385563
--- /dev/null
+++ b/studio/frontend/src/features/rag/components/project-sources-panel.tsx
@@ -0,0 +1,138 @@
+// SPD... | [
"studio/frontend/src/features/rag/components/project-sources-panel.tsx"
] | [
{
"comment": "**<sub><sub></sub></sub> Refresh source cache after uploads settle**\n\nWhen a user adds project sources and submits a chat before `upload(files)` has created the document row, `projectHasSources()` can observe an empty list and cache ... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | **<sub><sub></sub></sub> Validate project existence before ingesting sources**
Unlike knowledge-base uploads, this endpoint starts ingestion without checking that `project_id` still refers to an existing project. With a stale `/chat?project=...` bookmark o... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/backend/routes/rag.py b/studio/backend/routes/rag.py
index 8aba73caac..8d23240fd5 100644
--- a/studio/backend/routes/rag.py
+++ b/studio/backend/routes/rag.py
@@ -75,6 +75,19 @@ def _save_upload(file: UploadFile) -> tuple[str, str]:
return stored_path, filename
+def _remove_stored_upload(st... | [
"studio/backend/routes/rag.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Validate project existence before ingesting sources**\n\nUnlike knowledge-base uploads, this endpoint starts ingestion without checking that `project_id` still refers to an existing project. With a stale `/chat... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | **<sub><sub></sub></sub> Restore visible focus styles for keyboard navigation**
When users navigate with the keyboard, this global `:focus` rule removes both the browser outline and Tailwind ring from every button/link/summary/role-button, including contro... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/frontend/src/index.css b/studio/frontend/src/index.css
index 2dc5888062..9299545da9 100644
--- a/studio/frontend/src/index.css
+++ b/studio/frontend/src/index.css
@@ -1260,6 +1260,25 @@
box-shadow: 0 0 0 2px color-mix(in oklab, var(--ring) 45%, transparent);
}
+ /* Clickable controls never sh... | [
"studio/frontend/src/index.css"
] | [
{
"comment": "**<sub><sub></sub></sub> Restore visible focus styles for keyboard navigation**\n\nWhen users navigate with the keyboard, this global `:focus` rule removes both the browser outline and Tailwind ring from every button/link/summary/role-... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | **<sub><sub></sub></sub> Avoid scoping thread sends to stale projects**
When `resolvedThreadId` is a standalone thread, `resolveProjectId` still falls back to `activeProjectId` after reading `thread.projectId` as null. During navigation from a project to a... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/frontend/src/features/chat/api/chat-adapter.ts b/studio/frontend/src/features/chat/api/chat-adapter.ts
index 351ad5d5c7..7d9ee20a5f 100644
--- a/studio/frontend/src/features/chat/api/chat-adapter.ts
+++ b/studio/frontend/src/features/chat/api/chat-adapter.ts
@@ -2,6 +2,7 @@
// Copyright 2026-presen... | [
"studio/frontend/src/features/chat/api/chat-adapter.ts"
] | [
{
"comment": "**<sub><sub></sub></sub> Avoid scoping thread sends to stale projects**\n\nWhen `resolvedThreadId` is a standalone thread, `resolveProjectId` still falls back to `activeProjectId` after reading `thread.projectId` as null. During naviga... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | **<sub><sub></sub></sub> Unlink uploaded files when sources are removed**
When a user removes a project source from the new Sources tab, this path only calls `handleRemove`/`deleteDocument`, and the delete endpoint removes the document and index rows witho... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/frontend/src/features/rag/components/project-sources-panel.tsx b/studio/frontend/src/features/rag/components/project-sources-panel.tsx
new file mode 100644
index 0000000000..941e385563
--- /dev/null
+++ b/studio/frontend/src/features/rag/components/project-sources-panel.tsx
@@ -0,0 +1,138 @@
+// SPD... | [
"studio/frontend/src/features/rag/components/project-sources-panel.tsx"
] | [
{
"comment": "**<sub><sub></sub></sub> Unlink uploaded files when sources are removed**\n\nWhen a user removes a project source from the new Sources tab, this path only calls `handleRemove`/`deleteDocument`, and the delete endpoint removes the docum... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | **<sub><sub></sub></sub> Allow retrying failed project-source ingests**
If the first ingestion for a project source fails, retrying the same file through this new endpoint immediately hits `start_ingestion`'s content-hash dedupe for the existing document r... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/backend/routes/rag.py b/studio/backend/routes/rag.py
index 8aba73caac..8d23240fd5 100644
--- a/studio/backend/routes/rag.py
+++ b/studio/backend/routes/rag.py
@@ -75,6 +75,19 @@ def _save_upload(file: UploadFile) -> tuple[str, str]:
return stored_path, filename
+def _remove_stored_upload(st... | [
"studio/backend/routes/rag.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Allow retrying failed project-source ingests**\n\nIf the first ingestion for a project source fails, retrying the same file through this new endpoint immediately hits `start_ingestion`'s content-hash dedupe for... | true | ||
unslothai/unsloth | 6,205 | comment_to_fix | Studio: project sources backed by RAG | **<sub><sub></sub></sub> Remove duplicate project uploads after dedupe**
When a user re-adds the same project source (for example after reloading the page, or with the same content under another filename), this saves a fresh upload before `start_ingestion(... | ebe3bbd0412b36d775025f485884114230966fc6 | a73a851d92516d580ddbce36dfcc4939249e64a9 | diff --git a/studio/backend/routes/rag.py b/studio/backend/routes/rag.py
index 8aba73caac..8d23240fd5 100644
--- a/studio/backend/routes/rag.py
+++ b/studio/backend/routes/rag.py
@@ -75,6 +75,19 @@ def _save_upload(file: UploadFile) -> tuple[str, str]:
return stored_path, filename
+def _remove_stored_upload(st... | [
"studio/backend/routes/rag.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Remove duplicate project uploads after dedupe**\n\nWhen a user re-adds the same project source (for example after reloading the page, or with the same content under another filename), this saves a fresh upload ... | true | ||
unslothai/unsloth | 6,254 | issue_to_patch | Attach DiffusionGemma visual-server from the prebuilt bundle | ## Summary
A fresh Studio install resolves and downloads the published llama.cpp prebuilt bundle (currently `unslothai/llama.cpp@b9601-mix-a0e2906`), which ships `llama-diffusion-gemma-visual-server` alongside `llama-server`. The runtime overlay allowlist in `runtime_patterns_for_choice` only kept `llama-server`, `lla... | ce34aba932d9408fb32877e8f2deeb69933edca7 | 70b1fe92cdeedb15d80853cfda614fb5a2c6ba11 | diff --git a/studio/install_llama_prebuilt.py b/studio/install_llama_prebuilt.py
index 92d1f84b2d..b8e1cfe9c0 100644
--- a/studio/install_llama_prebuilt.py
+++ b/studio/install_llama_prebuilt.py
@@ -4768,8 +4768,10 @@ def runtime_patterns_for_choice(choice: AssetChoice) -> list[str]:
# repackage the SO/DLL set (e.... | [
"studio/install_llama_prebuilt.py",
"tests/studio/install/test_rocm_support.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Require visual server in matching installs**\n\nWhen a user already has this same prebuilt release installed from before this change, the metadata fingerprint and health checks still match because they do not i... | diff --git a/tests/studio/install/test_rocm_support.py b/tests/studio/install/test_rocm_support.py
index 04463fb720..1a7de1af6d 100644
--- a/tests/studio/install/test_rocm_support.py
+++ b/tests/studio/install/test_rocm_support.py
@@ -393,6 +393,19 @@ def test_macos_patterns(self):
patterns = runtime_patterns_... | true | |
unslothai/unsloth | 6,252 | issue_to_patch | Fix Studio S3 dataset panel layout | ## Summary
- increase the shared Studio config card height from 470px to 520px
- render S3 dataset settings inline instead of as a nested card inside Dataset
## Testing
- npm run build
- git diff --check
| 522f308be229cd26c67dc20514538eebf3998362 | 894789f9cd1de74dffbc997e36cdaeb2c6a11470 | diff --git a/studio/frontend/src/features/studio/sections/s3-config-form.tsx b/studio/frontend/src/features/studio/sections/s3-config-form.tsx
index 72af75db30..8126027cf9 100644
--- a/studio/frontend/src/features/studio/sections/s3-config-form.tsx
+++ b/studio/frontend/src/features/studio/sections/s3-config-form.tsx
@... | [
"studio/frontend/src/features/studio/sections/s3-config-form.tsx",
"studio/frontend/src/index.css"
] | [] | true | ||
unslothai/unsloth | 5,828 | issue_to_patch | Studio: cache MCP tool discovery instead of re-probing every chat send | Right now Studio re-fetches the tool list from every enabled MCP server each time you send a chat message, and it does this before the model starts replying. The tool list almost never changes during a session, so this is wasted work. With a slow server it's especially noticeable: every single message hangs for a few s... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/core/inference/mcp_client.py b/studio/backend/core/inference/mcp_client.py
index a0d79bbdf4..5a36d90c5d 100644
--- a/studio/backend/core/inference/mcp_client.py
+++ b/studio/backend/core/inference/mcp_client.py
@@ -8,6 +8,7 @@
import os
import shlex
import sys
+import time
from typing im... | [
"studio/backend/core/inference/mcp_client.py",
"studio/backend/core/inference/tools.py",
"studio/backend/routes/mcp_servers.py",
"studio/backend/tests/test_mcp_servers.py",
"studio/backend/tests/test_mcp_stdio_pr5863.py"
] | [
{
"comment": "\n\nIf an enabled MCP server is offline or unreachable, every chat message will trigger a probe that blocks the chat send for the duration of the timeout (`PROBE_TIMEOUT_SECONDS` = 8s, or `OAUTH_PROBE_TIMEOUT_SECONDS` = 305s!). Sinc... | diff --git a/studio/backend/tests/test_mcp_servers.py b/studio/backend/tests/test_mcp_servers.py
index ede3cf15d4..90b1ade03c 100644
--- a/studio/backend/tests/test_mcp_servers.py
+++ b/studio/backend/tests/test_mcp_servers.py
@@ -598,3 +598,614 @@ def fake_single_turn(messages):
)
# Empty allow-list = run an... | true | |
unslothai/unsloth | 5,828 | comment_to_fix | Studio: cache MCP tool discovery instead of re-probing every chat send | 
If an enabled MCP server is offline or unreachable, every chat message will trigger a probe that blocks the chat send for the duration of the timeout (`PROBE_TIMEOUT_SECONDS` = 8s, or `OAUTH_PROBE_TIMEOUT_SECONDS` = 305s!). Since failures are not cache... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/core/inference/mcp_client.py b/studio/backend/core/inference/mcp_client.py
index a0d79bbdf4..5a36d90c5d 100644
--- a/studio/backend/core/inference/mcp_client.py
+++ b/studio/backend/core/inference/mcp_client.py
@@ -8,6 +8,7 @@
import os
import shlex
import sys
+import time
from typing im... | [
"studio/backend/core/inference/mcp_client.py"
] | [
{
"comment": "\n\nIf an enabled MCP server is offline or unreachable, every chat message will trigger a probe that blocks the chat send for the duration of the timeout (`PROBE_TIMEOUT_SECONDS` = 8s, or `OAUTH_PROBE_TIMEOUT_SECONDS` = 305s!). Sinc... | true | ||
unslothai/unsloth | 5,828 | comment_to_fix | Studio: cache MCP tool discovery instead of re-probing every chat send | 
Import the new helper functions `get_last_failure_time` and `record_failure` from `mcp_client` to support the cool-off period for failed probes.
```suggestion
from core.inference.mcp_client import (
MCP_TOOL_PREFIX,
OAUTH_PROBE_TIMEOUT_SECONDS... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/core/inference/tools.py b/studio/backend/core/inference/tools.py
index b27fa6ff73..8c45706395 100644
--- a/studio/backend/core/inference/tools.py
+++ b/studio/backend/core/inference/tools.py
@@ -24,11 +24,16 @@
from core.inference.mcp_client import (
MCP_TOOL_PREFIX,
+ TOOL_CACHE_... | [
"studio/backend/core/inference/tools.py"
] | [
{
"comment": "\n\nImport the new helper functions `get_last_failure_time` and `record_failure` from `mcp_client` to support the cool-off period for failed probes.\n\n```suggestion\nfrom core.inference.mcp_client import (\n MCP_TOOL_PREFIX,\n ... | true | ||
unslothai/unsloth | 5,828 | comment_to_fix | Studio: cache MCP tool discovery instead of re-probing every chat send | 
Implement the 30-second cool-off period for failed probes. This ensures that if a server is offline or unreachable, we do not block the chat path with a multi-second (or multi-minute) timeout on every single message send.
```python
import time
... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/core/inference/tools.py b/studio/backend/core/inference/tools.py
index b27fa6ff73..8c45706395 100644
--- a/studio/backend/core/inference/tools.py
+++ b/studio/backend/core/inference/tools.py
@@ -24,11 +24,16 @@
from core.inference.mcp_client import (
MCP_TOOL_PREFIX,
+ TOOL_CACHE_... | [
"studio/backend/core/inference/tools.py"
] | [
{
"comment": "\n\nImplement the 30-second cool-off period for failed probes. This ensures that if a server is offline or unreachable, we do not block the chat path with a multi-second (or multi-minute) timeout on every single message send.\n\n```... | true | ||
unslothai/unsloth | 5,828 | comment_to_fix | Studio: cache MCP tool discovery instead of re-probing every chat send | 
Import `record_failure` from `mcp_client` to allow manual refresh failures to also trigger the cool-off period.
```python
from core.inference.mcp_client import (
OAUTH_PROBE_TIMEOUT_SECONDS,
PROBE_TIMEOUT_SECONDS,
TOOL_CACHE_INVALIDATING_F... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/routes/mcp_servers.py b/studio/backend/routes/mcp_servers.py
index 3001c6b7c9..37d99a222e 100644
--- a/studio/backend/routes/mcp_servers.py
+++ b/studio/backend/routes/mcp_servers.py
@@ -10,12 +10,16 @@
from auth.authentication import get_current_subject
from core.inference.mcp_client im... | [
"studio/backend/routes/mcp_servers.py"
] | [
{
"comment": "\n\nImport `record_failure` from `mcp_client` to allow manual refresh failures to also trigger the cool-off period.\n\n```python\nfrom core.inference.mcp_client import (\n OAUTH_PROBE_TIMEOUT_SECONDS,\n PROBE_TIMEOUT_SECONDS,\... | true | ||
unslothai/unsloth | 5,828 | comment_to_fix | Studio: cache MCP tool discovery instead of re-probing every chat send | 
Record the failure when a manual refresh fails. This ensures that if a server is offline and the user tries to refresh it, the chat path won't immediately hang on the next message send.
```suggestion
except Exception as exc: # noqa: BLE001 — surf... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/routes/mcp_servers.py b/studio/backend/routes/mcp_servers.py
index 3001c6b7c9..37d99a222e 100644
--- a/studio/backend/routes/mcp_servers.py
+++ b/studio/backend/routes/mcp_servers.py
@@ -10,12 +10,16 @@
from auth.authentication import get_current_subject
from core.inference.mcp_client im... | [
"studio/backend/routes/mcp_servers.py"
] | [
{
"comment": "\n\nRecord the failure when a manual refresh fails. This ensures that if a server is offline and the user tries to refresh it, the chat path won't immediately hang on the next message send.\n\n```suggestion\n except Exception as ... | true | ||
unslothai/unsloth | 5,828 | comment_to_fix | Studio: cache MCP tool discovery instead of re-probing every chat send | 
Update the test to mock `time.time` and handle the 30-second cool-off period for failed probes.
```python
def test_get_enabled_mcp_tools_does_not_cache_failures(tmp_path, monkeypatch):
"""A failed probe must not be cached, so a recovered server is... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/tests/test_mcp_servers.py b/studio/backend/tests/test_mcp_servers.py
index ede3cf15d4..90b1ade03c 100644
--- a/studio/backend/tests/test_mcp_servers.py
+++ b/studio/backend/tests/test_mcp_servers.py
@@ -598,3 +598,614 @@ def fake_single_turn(messages):
)
# Empty allow-list = run an... | [
"studio/backend/tests/test_mcp_servers.py"
] | [
{
"comment": "\n\nUpdate the test to mock `time.time` and handle the 30-second cool-off period for failed probes.\n\n```python\ndef test_get_enabled_mcp_tools_does_not_cache_failures(tmp_path, monkeypatch):\n \"\"\"A failed probe must not be c... | true | ||
unslothai/unsloth | 5,828 | comment_to_fix | Studio: cache MCP tool discovery instead of re-probing every chat send | 
Add a new unit test to verify that the cool-off period correctly prevents immediate re-probing of failed servers.
```python
def test_get_enabled_mcp_tools_skips_cache_when_config_changes_mid_probe(
tmp_path, monkeypatch
):
"""A config edit lan... | aefe904d66ec026a559812048436c9db7dd4e557 | d0163348490c990324ecf8e7316d32b73c707b88 | diff --git a/studio/backend/tests/test_mcp_servers.py b/studio/backend/tests/test_mcp_servers.py
index ede3cf15d4..90b1ade03c 100644
--- a/studio/backend/tests/test_mcp_servers.py
+++ b/studio/backend/tests/test_mcp_servers.py
@@ -598,3 +598,614 @@ def fake_single_turn(messages):
)
# Empty allow-list = run an... | [
"studio/backend/tests/test_mcp_servers.py"
] | [
{
"comment": "\n\nAdd a new unit test to verify that the cool-off period correctly prevents immediate re-probing of failed servers.\n\n```python\ndef test_get_enabled_mcp_tools_skips_cache_when_config_changes_mid_probe(\n tmp_path, monkeypatch... | true | ||
unslothai/unsloth | 6,245 | issue_to_patch | Studio: bundle Gemma 4 chat templates (E2B/E4B + larger) and auto-apply to unsloth/gemma-4-*-GGUF | ## What
Studio now ships bundled Gemma 4 chat templates and applies them to `unsloth/gemma-4-*-GGUF` models at llama-server launch via `--chat-template-file`. The templates are based on Google's Gemma 4 chat-template update (HF `google/gemma-4-31B-it` discussion #118), which adds the `preserve_thinking` flag. We defau... | cd8e6c1e7709364e33ee2ecc56841d5f9b60c7da | 8b84fcdd8449e56d7b3a5f57935d968d0c2de5f8 | diff --git a/pyproject.toml b/pyproject.toml
index 16fd9f833f..e807318ce5 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -55,6 +55,7 @@ studio = [
"frontend/.git*",
"backend/requirements/**/*",
"backend/plugins/**/*",
+ "backend/assets/**/*.jinja",
"backend/core/data_recipe/oxc-validator/*.j... | [
"pyproject.toml",
"studio/backend/assets/chat_templates/gemma-4-edge.jinja",
"studio/backend/assets/chat_templates/gemma-4.jinja",
"studio/backend/core/inference/chat_templates.py",
"studio/backend/core/inference/llama_cpp.py",
"studio/backend/routes/inference.py",
"studio/backend/tests/test_gemma4_chat... | [
{
"comment": "\n\nModifying `sys.modules` globally using `setdefault` pollutes the global module registry for the entire test session. This can lead to flaky tests or unexpected failures in other test files that run after this one and rely on... | diff --git a/studio/backend/tests/test_gemma4_chat_template_override.py b/studio/backend/tests/test_gemma4_chat_template_override.py
new file mode 100644
index 0000000000..f726741aa5
--- /dev/null
+++ b/studio/backend/tests/test_gemma4_chat_template_override.py
@@ -0,0 +1,345 @@
+# SPDX-License-Identifier: AGPL-3.0-onl... | true | |
unslothai/unsloth | 6,250 | issue_to_patch | Studio: serve DiffusionGemma with live in-place denoising and honest stats | Adds first-class DiffusionGemma support to Unsloth Studio. A block-diffusion GGUF (a diffusion architecture, or a `diffusion.canvas_length` KV) is served by an on-device visual decoder instead of llama-server, behind the same OpenAI-compatible interface, so the rest of Studio is unchanged.
## Backend
- Detects block-... | cd8e6c1e7709364e33ee2ecc56841d5f9b60c7da | 8ad878b30453e475c6dd1722de0ce782a64b0f85 | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index e2de5fc322..486cff1bc8 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -722,6 +722,11 @@ def __init__(self):
self._spec_fallback_reason: Optiona... | [
"studio/backend/core/inference/llama_cpp.py",
"studio/backend/models/inference.py",
"studio/backend/routes/inference.py",
"studio/frontend/src/components/assistant-ui/markdown-text.tsx",
"studio/frontend/src/components/assistant-ui/message-timing.tsx",
"studio/frontend/src/components/assistant-ui/thread.t... | [
{
"comment": "\n\nOn Windows, the visual-server binary is installed with a `.exe` extension (e.g., `llama-diffusion-gemma-visual-server.exe`). However, `_find_diffusion_assets` looks for the extensionless name `llama-diffusion-gemma-visual-server... | true | ||
unslothai/unsloth | 6,251 | issue_to_patch | Studio: refine menu chevron, tick icon, and one-line plus-menu shape | Two small Studio menu polish tweaks.
## Submenu chevron sizing
The `>` chevron on submenu triggers was sized at `size-4` (16px), which read a touch heavy next to the row text. Dropped it to `size-[12px]` so it sits lighter while staying vertically centered via the row's `items-center` and `ml-auto`. Done centrally in ... | f4fc06b5bb4f6174e8f39599ba1fb5d8b054e48d | 62128927c43d20f47a51cd9977aefae96de58ac9 | diff --git a/studio/frontend/src/components/ui/context-menu.tsx b/studio/frontend/src/components/ui/context-menu.tsx
index 9fda100bf1..fbe0c2c741 100644
--- a/studio/frontend/src/components/ui/context-menu.tsx
+++ b/studio/frontend/src/components/ui/context-menu.tsx
@@ -127,7 +127,7 @@ function ContextMenuSubTrigger({
... | [
"studio/frontend/src/components/ui/context-menu.tsx",
"studio/frontend/src/components/ui/dropdown-menu.tsx",
"studio/frontend/src/components/ui/menubar.tsx",
"studio/frontend/src/index.css"
] | [
{
"comment": "**<sub><sub></sub></sub> Do not vendor the Pro-only icon path**\n\nThis new shared icon copies the `duotone-standard` Hugeicons geometry directly into the AGPL source, while the repo only declares the free `@hugeicons/core-free-icons` ... | true | ||
unslothai/unsloth | 6,169 | issue_to_patch | Studio: persist speculative decoding preference across restart and model switch | The chat Speculative Decoding setting didn't persist. It reset to Auto on every model switch and on app restart, so a user who set it to Off or Ngram lost that choice constantly. Its sibling load settings (sampling, context length, system prompt) do persist.
The fix treats it like those other standing preferences: p... | 911ceba7faedc0e4ff7db23cd0941aebf793952d | 7d5dcbb4d0fc9fba165edcc7312fe82d0f7f81ed | diff --git a/studio/frontend/src/features/chat/api/chat-adapter.ts b/studio/frontend/src/features/chat/api/chat-adapter.ts
index 5903bef107..824c8dcbd8 100644
--- a/studio/frontend/src/features/chat/api/chat-adapter.ts
+++ b/studio/frontend/src/features/chat/api/chat-adapter.ts
@@ -36,7 +36,10 @@ import {
import {
... | [
"studio/frontend/src/features/chat/api/chat-adapter.ts",
"studio/frontend/src/features/chat/hooks/use-chat-model-runtime.ts",
"studio/frontend/src/features/chat/lib/apply-inference-status-to-store.ts",
"studio/frontend/src/features/chat/shared-composer.tsx",
"studio/frontend/src/features/chat/stores/chat-ru... | [
{
"comment": "**<sub><sub></sub></sub> Don't mark the persisted spec mode as loaded before load succeeds**\n\nIf a model switch fails after the old model is unloaded, the catch block reloads the previous model without passing `speculative_type` and ... | true | ||
unslothai/unsloth | 6,169 | comment_to_fix | Studio: persist speculative decoding preference across restart and model switch | **<sub><sub></sub></sub> Don't mark the persisted spec mode as loaded before load succeeds**
If a model switch fails after the old model is unloaded, the catch block reloads the previous model without passing `speculative_type` and then calls `refresh()`. ... | 911ceba7faedc0e4ff7db23cd0941aebf793952d | 7d5dcbb4d0fc9fba165edcc7312fe82d0f7f81ed | diff --git a/studio/frontend/src/features/chat/hooks/use-chat-model-runtime.ts b/studio/frontend/src/features/chat/hooks/use-chat-model-runtime.ts
index 70cba21afc..e513d351c4 100644
--- a/studio/frontend/src/features/chat/hooks/use-chat-model-runtime.ts
+++ b/studio/frontend/src/features/chat/hooks/use-chat-model-runt... | [
"studio/frontend/src/features/chat/hooks/use-chat-model-runtime.ts"
] | [
{
"comment": "**<sub><sub></sub></sub> Don't mark the persisted spec mode as loaded before load succeeds**\n\nIf a model switch fails after the old model is unloaded, the catch block reloads the previous model without passing `speculative_type` and ... | true | ||
unslothai/unsloth | 6,169 | comment_to_fix | Studio: persist speculative decoding preference across restart and model switch | **<sub><sub></sub></sub> Avoid persisting unapplied speculative edits**
Because `setSpeculativeType` now writes localStorage immediately, changing the load setting and then pressing Reset can still leave the canceled value persisted. For example, with Auto... | 911ceba7faedc0e4ff7db23cd0941aebf793952d | 7d5dcbb4d0fc9fba165edcc7312fe82d0f7f81ed | diff --git a/studio/frontend/src/features/chat/stores/chat-runtime-store.ts b/studio/frontend/src/features/chat/stores/chat-runtime-store.ts
index 2d23777378..7aecc1fcd9 100644
--- a/studio/frontend/src/features/chat/stores/chat-runtime-store.ts
+++ b/studio/frontend/src/features/chat/stores/chat-runtime-store.ts
@@ -4... | [
"studio/frontend/src/features/chat/stores/chat-runtime-store.ts"
] | [
{
"comment": "**<sub><sub></sub></sub> Avoid persisting unapplied speculative edits**\n\nBecause `setSpeculativeType` now writes localStorage immediately, changing the load setting and then pressing Reset can still leave the canceled value persisted... | true | ||
unslothai/unsloth | 6,169 | comment_to_fix | Studio: persist speculative decoding preference across restart and model switch | **<sub><sub></sub></sub> Reset session-only spec mode before compare loads**
In generalized compare mode, this load path bypasses the normal `selectModel` switch logic that resets session-only MTP modes back to the persisted universal preference. If the cu... | 911ceba7faedc0e4ff7db23cd0941aebf793952d | 7d5dcbb4d0fc9fba165edcc7312fe82d0f7f81ed | diff --git a/studio/frontend/src/features/chat/shared-composer.tsx b/studio/frontend/src/features/chat/shared-composer.tsx
index 9f190355de..a8fa2360a0 100644
--- a/studio/frontend/src/features/chat/shared-composer.tsx
+++ b/studio/frontend/src/features/chat/shared-composer.tsx
@@ -75,6 +75,9 @@ import {
} from "./sto... | [
"studio/frontend/src/features/chat/shared-composer.tsx"
] | [
{
"comment": "**<sub><sub></sub></sub> Reset session-only spec mode before compare loads**\n\nIn generalized compare mode, this load path bypasses the normal `selectModel` switch logic that resets session-only MTP modes back to the persisted univers... | true | ||
unslothai/unsloth | 5,771 | issue_to_patch | Studio: new-chat shortcut, composer draft autosave, archive threads | ## Summary
- Bind **Cmd/Ctrl + Shift + O** to open a new chat from anywhere in Studio
- Persist per-thread composer drafts in `localStorage` so a half-typed message survives a
navigation or reload
- Add **Archive** / **Unarchive** actions to chat threads, plus a collapsible **Archived** group
in the si... | 3427e3fd6236873d07cfa334778b6384541e8868 | fa5456ba48ad5d78686e62dde7f272de6f25c350 | diff --git a/studio/frontend/src/app/routes/__root.tsx b/studio/frontend/src/app/routes/__root.tsx
index c05b0e6451..f5179e68f7 100644
--- a/studio/frontend/src/app/routes/__root.tsx
+++ b/studio/frontend/src/app/routes/__root.tsx
@@ -6,7 +6,7 @@ import { Navbar } from "@/components/navbar";
import { fetchDeviceType, ... | [
"studio/frontend/src/app/routes/__root.tsx",
"studio/frontend/src/components/app-sidebar.tsx",
"studio/frontend/src/components/assistant-ui/thread.tsx",
"studio/frontend/src/features/chat/hooks/use-chat-sidebar-items.ts",
"studio/frontend/src/features/chat/index.ts",
"studio/frontend/src/features/chat/uti... | [
{
"comment": "\n\nThere is a race condition when switching threads. When `activeThreadId` changes, `draftKey` updates immediately, but `composerText` (from `useAuiState`) may still hold the value from the previous thread for one render cycle. Thi... | true | ||
unslothai/unsloth | 5,771 | comment_to_fix | Studio: new-chat shortcut, composer draft autosave, archive threads | 
There is a race condition when switching threads. When `activeThreadId` changes, `draftKey` updates immediately, but `composerText` (from `useAuiState`) may still hold the value from the previous thread for one render cycle. This results in the previou... | 3427e3fd6236873d07cfa334778b6384541e8868 | fa5456ba48ad5d78686e62dde7f272de6f25c350 | diff --git a/studio/frontend/src/components/assistant-ui/thread.tsx b/studio/frontend/src/components/assistant-ui/thread.tsx
index 9ff6f440b3..ac5ae19aac 100644
--- a/studio/frontend/src/components/assistant-ui/thread.tsx
+++ b/studio/frontend/src/components/assistant-ui/thread.tsx
@@ -71,8 +71,11 @@ import { useChatRu... | [
"studio/frontend/src/components/assistant-ui/thread.tsx"
] | [
{
"comment": "\n\nThere is a race condition when switching threads. When `activeThreadId` changes, `draftKey` updates immediately, but `composerText` (from `useAuiState`) may still hold the value from the previous thread for one render cycle. Thi... | true | ||
unslothai/unsloth | 5,771 | comment_to_fix | Studio: new-chat shortcut, composer draft autosave, archive threads | 
If a thread has no saved draft, this effect returns early without updating the composer. Since the composer state is shared within the runtime, switching from a thread with a draft to one without a draft will leave the previous thread's text in the... | 3427e3fd6236873d07cfa334778b6384541e8868 | fa5456ba48ad5d78686e62dde7f272de6f25c350 | diff --git a/studio/frontend/src/components/assistant-ui/thread.tsx b/studio/frontend/src/components/assistant-ui/thread.tsx
index 9ff6f440b3..ac5ae19aac 100644
--- a/studio/frontend/src/components/assistant-ui/thread.tsx
+++ b/studio/frontend/src/components/assistant-ui/thread.tsx
@@ -71,8 +71,11 @@ import { useChatRu... | [
"studio/frontend/src/components/assistant-ui/thread.tsx"
] | [
{
"comment": "\n\nIf a thread has no saved draft, this effect returns early without updating the composer. Since the composer state is shared within the runtime, switching from a thread with a draft to one without a draft will leave the previ... | true | ||
unslothai/unsloth | 5,771 | comment_to_fix | Studio: new-chat shortcut, composer draft autosave, archive threads | 
The dropdown menu for archived threads is missing the "Rename" action. For consistency with the "Recents" section and better user experience, users should be able to rename threads even after they have been archived. It is acceptable to keep this c... | 3427e3fd6236873d07cfa334778b6384541e8868 | fa5456ba48ad5d78686e62dde7f272de6f25c350 | diff --git a/studio/frontend/src/components/app-sidebar.tsx b/studio/frontend/src/components/app-sidebar.tsx
index 72bb75a84c..16ab7932b1 100644
--- a/studio/frontend/src/components/app-sidebar.tsx
+++ b/studio/frontend/src/components/app-sidebar.tsx
@@ -45,6 +45,8 @@ import { Switch } from "@/components/ui/switch";
i... | [
"studio/frontend/src/components/app-sidebar.tsx"
] | [
{
"comment": "\n\nThe dropdown menu for archived threads is missing the \"Rename\" action. For consistency with the \"Recents\" section and better user experience, users should be able to rename threads even after they have been archived. It ... | true | ||
unslothai/unsloth | 6,247 | issue_to_patch | Studio: fix Downloaded model list disappearing and order it by last download | ## Summary
In the chat model picker ("Search Model"), the Downloaded section sometimes did not appear, and already downloaded models showed up under Recommended with their quants labelled "downloaded". The Downloaded list was also unordered. This change fixes the root cause, orders Downloaded by last download, and kee... | 40cb0a71b4ea59e0b54b0932860f6f1eafb58570 | e930e774245f37d6c08f48ee9f06f2555aa79702 | diff --git a/studio/backend/hub/utils/inventory_scan.py b/studio/backend/hub/utils/inventory_scan.py
index fe3aadfd38..0f7ce6fe34 100644
--- a/studio/backend/hub/utils/inventory_scan.py
+++ b/studio/backend/hub/utils/inventory_scan.py
@@ -142,13 +142,18 @@ def _compute_all_hf_cache_scans() -> list:
logger.warn... | [
"studio/backend/hub/utils/inventory_scan.py",
"studio/backend/routes/models.py",
"studio/backend/tests/test_cached_gguf_routes.py",
"studio/backend/utils/models/model_config.py",
"studio/frontend/src/components/assistant-ui/model-selector/pickers.tsx",
"studio/frontend/src/features/chat/api/chat-api.ts"
] | [
{
"comment": "\n\nWhile the rest of the cache scanning logic has been robustly guarded against inaccessible directories and broken symlinks, `f.stat().st_size` inside `get_gguf_variants` remains unguarded. If any snapshot contains a broken sy... | diff --git a/studio/backend/tests/test_cached_gguf_routes.py b/studio/backend/tests/test_cached_gguf_routes.py
index 20aa37a2b4..5a4ca68ab4 100644
--- a/studio/backend/tests/test_cached_gguf_routes.py
+++ b/studio/backend/tests/test_cached_gguf_routes.py
@@ -371,3 +371,156 @@ def test_list_cached_gguf_includes_vision_r... | true | |
unslothai/unsloth | 6,249 | issue_to_patch | Studio: use px-2.5 for Hub option menu padding | ### What this does
Follow-up to #6248. Swaps the one arbitrary `px-[10px]` on the Hub option menu surface for the standard `px-2.5` utility, addressing the review note on that PR.
Both resolve to exactly `10px` (`0.625rem` at the default 16px root), so this is a no-op visually and only keeps the menu padding on the s... | 9df607c0dee620ea88f73b79a1fa79ec7871a308 | c0ae8cfa90ee27ab09f72ee2d351fba089fa8936 | diff --git a/studio/frontend/src/features/hub/catalog/hub-option-menu.tsx b/studio/frontend/src/features/hub/catalog/hub-option-menu.tsx
index a1542ef5c8..5d7532d2c0 100644
--- a/studio/frontend/src/features/hub/catalog/hub-option-menu.tsx
+++ b/studio/frontend/src/features/hub/catalog/hub-option-menu.tsx
@@ -185,7 +18... | [
"studio/frontend/src/features/hub/catalog/hub-option-menu.tsx"
] | [] | true | ||
unslothai/unsloth | 6,248 | issue_to_patch | Studio: fully rounded Hub pills and refreshed menu icons | ### What this does
Rounds out the Hub chrome so the pills read consistently, aligns the All formats style dropdown with the composer "+" menu, widens the chat options menu so "Move to project" fits on one line, and swaps a few icons to their HugeIcons equivalents.
### Hub pills and dropdown
- `hub.css`: `hub-tag-sof... | 240c0c350027e74dfef00a44eead7154768705fb | a79d934749cbd04213ab34d076da864169cd3474 | diff --git a/studio/frontend/src/components/app-sidebar.tsx b/studio/frontend/src/components/app-sidebar.tsx
index 41d1b54232..72bb75a84c 100644
--- a/studio/frontend/src/components/app-sidebar.tsx
+++ b/studio/frontend/src/components/app-sidebar.tsx
@@ -610,7 +610,7 @@ export function AppSidebar() {
side=... | [
"studio/frontend/src/components/app-sidebar.tsx",
"studio/frontend/src/components/assistant-ui/model-selector/pickers.tsx",
"studio/frontend/src/components/ui/command.tsx",
"studio/frontend/src/features/chat/components/chat-search-dialog.tsx",
"studio/frontend/src/features/chat/prompt-storage/prompt-storage... | [
{
"comment": "\n\nTo maintain consistency with the rest of the codebase and adhere to standard Tailwind CSS spacing utilities, prefer using `px-2.5` instead of the arbitrary value `px-[10px]`. Both resolve to exactly `10px` (`0.625rem`).\n\n`... | true | ||
unslothai/unsloth | 6,227 | issue_to_patch | Studio: fix WSL Strix Halo GPU on reinstall (ROCDXG drop-in + system HIP before bundle) | ## Summary
Three fixes that restore **gfx1151 GPU on WSL** for Unsloth Studio after PR #5940 landed. All are strictly WSL-gated (no-op on bare-metal Linux, NVIDIA, macOS, Windows). Found while re-running the full teardown→reinstall→verify cycle on a real Strix Halo (Radeon 8060S) under WSL2.
### 1. `install.sh`: pers... | 6dae2f525b5218143f85cc8f401313e182bccac3 | b149eb70975dc0be33ecb4f1e303c35e9c1741e2 | diff --git a/install.sh b/install.sh
index 532ac61bc0..2b8a5bb7d3 100755
--- a/install.sh
+++ b/install.sh
@@ -2043,6 +2043,37 @@ _pick_radeon_wheel() {
# CPU, non-Strix WSL) skips it and normal detection runs unchanged. NEVER aborts
# the installer -- always returns 0. Runs the idempotent helper (ROCm 7.2 +
# libro... | [
"install.sh",
"studio/backend/core/inference/llama_cpp.py",
"studio/install_llama_prebuilt.py",
"tests/studio/install/test_rocm_support.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Include lib64 when persisting the ROCm bridge path**\n\nWhen the guard above is satisfied by `/opt/rocm/lib64/librocdxg.so`, the drop-in still persists only `/opt/rocm/lib` in `LD_LIBRARY_PATH`. In that lib64-o... | diff --git a/tests/studio/install/test_rocm_support.py b/tests/studio/install/test_rocm_support.py
index 9aa06b6ff7..04463fb720 100644
--- a/tests/studio/install/test_rocm_support.py
+++ b/tests/studio/install/test_rocm_support.py
@@ -11,7 +11,7 @@
import subprocess
import sys
from pathlib import Path
-from unittest... | true | |
unslothai/unsloth | 6,202 | issue_to_patch | fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request | ## Problem
When calling `/v1/responses` with `chat_template_kwargs` (e.g. `{"enable_thinking": true}`), the flag is silently ignored — the model's load-time default reasoning mode is used instead. This breaks reasoning control for any OpenAI-Responses-SDK client (issue #6198).
## Root cause
`_build_chat_request` tra... | aba21db4665015111d61b06b8f1ee46b9a36fa13 | 58edcbd257607fcc2c1ef12b5410f82b163d60a8 | diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 7fa117980b..3cc7514ff5 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -990,8 +990,10 @@ def _request_reasoning_kwargs(
# enable_thinking / reas... | [
"studio/backend/core/inference/llama_cpp.py",
"studio/backend/models/inference.py",
"studio/backend/routes/inference.py",
"studio/backend/tests/test_openai_tool_passthrough.py",
"studio/backend/tests/test_responses_tool_passthrough.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Preserve `reasoning.none` through GGUF passthrough**\n\nWhen a Responses request sets `reasoning={\"effort\":\"none\"}` and the loaded GGUF template uses the `reasoning_effort` style (for example Harmony/gpt-os... | diff --git a/studio/backend/tests/test_openai_tool_passthrough.py b/studio/backend/tests/test_openai_tool_passthrough.py
index 79e8c9ae49..baa3e50b95 100644
--- a/studio/backend/tests/test_openai_tool_passthrough.py
+++ b/studio/backend/tests/test_openai_tool_passthrough.py
@@ -28,11 +28,13 @@
ChatMessage,
Co... | true | |
unslothai/unsloth | 6,202 | comment_to_fix | fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request | **<sub><sub></sub></sub> Preserve `reasoning.none` through GGUF passthrough**
When a Responses request sets `reasoning={"effort":"none"}` and the loaded GGUF template uses the `reasoning_effort` style (for example Harmony/gpt-oss), this sets `enable_thinki... | aba21db4665015111d61b06b8f1ee46b9a36fa13 | 58edcbd257607fcc2c1ef12b5410f82b163d60a8 | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 7b17e450d1..21390d7a3e 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -682,6 +682,8 @@ def _classify_llama_generation_error(exc: Exception) -> Optional[bool]:
ResponsesFunctionCal... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Preserve `reasoning.none` through GGUF passthrough**\n\nWhen a Responses request sets `reasoning={\"effort\":\"none\"}` and the loaded GGUF template uses the `reasoning_effort` style (for example Harmony/gpt-os... | true | ||
unslothai/unsloth | 6,202 | comment_to_fix | fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request | **<sub><sub></sub></sub> Do not upgrade minimal reasoning to high**
For local GGUF models that use `reasoning_effort` templates, a Responses request with `reasoning: {"effort": "minimal"}` enters this branch and also sets `enable_thinking=True`; later `Lla... | aba21db4665015111d61b06b8f1ee46b9a36fa13 | 58edcbd257607fcc2c1ef12b5410f82b163d60a8 | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 7b17e450d1..21390d7a3e 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -682,6 +682,8 @@ def _classify_llama_generation_error(exc: Exception) -> Optional[bool]:
ResponsesFunctionCal... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Do not upgrade minimal reasoning to high**\n\nFor local GGUF models that use `reasoning_effort` templates, a Responses request with `reasoning: {\"effort\": \"minimal\"}` enters this branch and also sets `enabl... | true | ||
unslothai/unsloth | 6,202 | comment_to_fix | fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request | **<sub><sub></sub></sub> Preserve literal think tags in visible output**
Because this parser runs on every Responses adapter output, an ordinary answer that legitimately contains literal `<think>...</think>` markup (for example when the user asks for XML/H... | aba21db4665015111d61b06b8f1ee46b9a36fa13 | 58edcbd257607fcc2c1ef12b5410f82b163d60a8 | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 7b17e450d1..21390d7a3e 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -682,6 +682,8 @@ def _classify_llama_generation_error(exc: Exception) -> Optional[bool]:
ResponsesFunctionCal... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Preserve literal think tags in visible output**\n\nBecause this parser runs on every Responses adapter output, an ordinary answer that legitimately contains literal `<think>...</think>` markup (for example when... | true | ||
unslothai/unsloth | 6,202 | comment_to_fix | fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request | **<sub><sub></sub></sub> Honor explicit thinking override before forwarding effort**
When a Responses request includes `chat_template_kwargs: {"enable_thinking": false}` together with `reasoning: {"effort": "high"}`, this still stores `reasoning_effort="hi... | aba21db4665015111d61b06b8f1ee46b9a36fa13 | 58edcbd257607fcc2c1ef12b5410f82b163d60a8 | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 7b17e450d1..21390d7a3e 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -682,6 +682,8 @@ def _classify_llama_generation_error(exc: Exception) -> Optional[bool]:
ResponsesFunctionCal... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Honor explicit thinking override before forwarding effort**\n\nWhen a Responses request includes `chat_template_kwargs: {\"enable_thinking\": false}` together with `reasoning: {\"effort\": \"high\"}`, this stil... | true | ||
unslothai/unsloth | 6,202 | comment_to_fix | fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request | **<sub><sub></sub></sub> Preserve reasoning-only replies as visible text**
When a local thinking model returns only `reasoning_content` / a `<think>...</think>` block and no visible `content` (the existing llama_cpp stream path explicitly handles this for ... | aba21db4665015111d61b06b8f1ee46b9a36fa13 | 58edcbd257607fcc2c1ef12b5410f82b163d60a8 | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 7b17e450d1..21390d7a3e 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -682,6 +682,8 @@ def _classify_llama_generation_error(exc: Exception) -> Optional[bool]:
ResponsesFunctionCal... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Preserve reasoning-only replies as visible text**\n\nWhen a local thinking model returns only `reasoning_content` / a `<think>...</think>` block and no visible `content` (the existing llama_cpp stream path expl... | true | ||
unslothai/unsloth | 6,202 | comment_to_fix | fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request | **<sub><sub></sub></sub> Gate think parsing on reasoning-capable GGUFs**
When a loaded GGUF does not advertise `supports_reasoning`, `LlamaCppBackend._request_reasoning_kwargs` sends no `chat_template_kwargs`, so a Responses request that includes `reasonin... | aba21db4665015111d61b06b8f1ee46b9a36fa13 | 58edcbd257607fcc2c1ef12b5410f82b163d60a8 | diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 7b17e450d1..21390d7a3e 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -682,6 +682,8 @@ def _classify_llama_generation_error(exc: Exception) -> Optional[bool]:
ResponsesFunctionCal... | [
"studio/backend/routes/inference.py"
] | [
{
"comment": "**<sub><sub></sub></sub> Gate think parsing on reasoning-capable GGUFs**\n\nWhen a loaded GGUF does not advertise `supports_reasoning`, `LlamaCppBackend._request_reasoning_kwargs` sends no `chat_template_kwargs`, so a Responses request... | true | ||
unslothai/unsloth | 6,233 | issue_to_patch | Studio: bottom update banners, smooth llama.cpp progress, re-prompt after copy | ## Summary
Three related improvements to the Studio update banners, following the llama.cpp banner design from #6196.
### Web "New Unsloth version" banner
- Restyled to match the llama.cpp banner: borderless 24px rounded card, drop shadow, and heading title.
- Moved from the top-right to the bottom-right, stacked wi... | c836228de8c7e26273e027d64dfdb37859180271 | f3faa54b983ce5ccba0858c76ce86af2c3d9a063 | diff --git a/studio/frontend/src/app/provider.tsx b/studio/frontend/src/app/provider.tsx
index 91d99e238e..f806a0f405 100644
--- a/studio/frontend/src/app/provider.tsx
+++ b/studio/frontend/src/app/provider.tsx
@@ -259,8 +259,16 @@ function TauriWrapper({ children }: { children: ReactNode }) {
<>
{child... | [
"studio/frontend/src/app/provider.tsx",
"studio/frontend/src/components/llama-update-banner.tsx",
"studio/frontend/src/components/web/update-banner.tsx",
"studio/frontend/src/hooks/use-llama-update-check.ts",
"studio/frontend/src/hooks/use-web-update-check.ts"
] | [
{
"comment": "\n\nIn some environments or browsers, the timestamp passed to `requestAnimationFrame` (`now`) can be slightly behind `performance.now()` when the effect is first initialized. This can result in `now < last`, leading to a negativ... | true | ||
unslothai/unsloth | 6,237 | issue_to_patch | Studio: pinnable plus menu items and saved prompt pins | ### What this does
The chat composer plus menu currently has a fixed layout. This makes the non-core items pinnable from Settings -> Chat: pinned items show at the top level of the menu and unpinned items move into the "More" submenu, which hides itself when it has no children. Individual saved prompts can also be pin... | b3ffa266dabefa0a50d872505ddb0b7ade3de6b1 | d3d9ebb7789be507062024e05fab85b380c9412f | diff --git a/studio/frontend/src/components/assistant-ui/thread.tsx b/studio/frontend/src/components/assistant-ui/thread.tsx
index 100143ed33..9ff6f440b3 100644
--- a/studio/frontend/src/components/assistant-ui/thread.tsx
+++ b/studio/frontend/src/components/assistant-ui/thread.tsx
@@ -69,6 +69,11 @@ import { getExtern... | [
"studio/frontend/src/components/assistant-ui/thread.tsx",
"studio/frontend/src/features/chat/index.ts",
"studio/frontend/src/features/chat/prompt-storage/prompt-storage-dialog.tsx",
"studio/frontend/src/features/chat/shared-composer.tsx",
"studio/frontend/src/features/chat/stores/plus-menu-prefs-store.ts",
... | [
{
"comment": " \n\nIt appears that the tool confirmation wrappers (`withToolConfirmation`) were accidentally removed from the tool definitions in `AssistantMessage`... | true | ||
unslothai/unsloth | 6,242 | issue_to_patch | Studio: login card polish and sidebar label alignment | ### What this does
Two small visual fixes: the login card gets slightly rounder corners, more padding, and a submit button that sizes to its label, and the sidebar rows shift 2px left with the section headers brought flush to the row text, like Gemini's sidebar.
### Changes
- `features/auth/login-page.tsx`: card rad... | 40cb0a71b4ea59e0b54b0932860f6f1eafb58570 | 7d70653ec43ea6bd206de8543ecadcc2a342b8f2 | diff --git a/studio/frontend/src/components/app-sidebar.tsx b/studio/frontend/src/components/app-sidebar.tsx
index faa98abeea..41d1b54232 100644
--- a/studio/frontend/src/components/app-sidebar.tsx
+++ b/studio/frontend/src/components/app-sidebar.tsx
@@ -949,7 +949,7 @@ export function AppSidebar() {
<... | [
"studio/frontend/src/components/app-sidebar.tsx",
"studio/frontend/src/features/auth/components/auth-form.tsx",
"studio/frontend/src/features/auth/login-page.tsx",
"studio/frontend/src/index.css"
] | [
{
"comment": "\n\nTo ensure compatibility with Right-to-Left (RTL) languages, prefer using logical padding classes (`ps-` and `pe-`) instead of physical padding classes (`pl-` and `pr-`). This ensures that the padding is applied to the correc... | true | ||
upstash/context7 | 2,763 | issue_to_patch | chore(release): version packages | This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to master, this PR wil... | 8123b519edacc5369bb71d4e067071949c51b9ac | f3d1d15db0f7f8f2cb5cc23d4d05324eaec0d045 | diff --git a/.changeset/node18-context7-mcp.md b/.changeset/node18-context7-mcp.md
deleted file mode 100644
index 3b2940d2d..000000000
--- a/.changeset/node18-context7-mcp.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@upstash/context7-mcp": patch
----
-
-Restore Node 18 support by pinning undici to ^6.26.0 and commander to ... | [
".changeset/node18-context7-mcp.md",
"packages/mcp/CHANGELOG.md",
"packages/mcp/package.json"
] | [] | true | ||
upstash/context7 | 2,762 | issue_to_patch | [Bug]: Stopped working for Node 18
### MCP Client
Claude Code
### Context7 MCP Version
latest
### Bug Description
Claude code plugin with Context7 stopped working. After some debugging I can see that some of updated deps now required node 20+.
```
npm warn EBADENGINE Unsupported engine {
npm warn EBADENGINE pa... | fix: restore Node 18 support for context7 mcp | ## Summary
- pin @upstash/context7-mcp runtime dependencies to Node 18-compatible major versions
- update the lockfile so Node 18 does not load undici 7 at startup
Closes #2761
## Validation
- pnpm install --filter @upstash/context7-mcp --frozen-lockfile
- pnpm --filter @upstash/context7-mcp build
- pnpm --filter @up... | 574fc642ab4ba6e178b56f436928cb564349cf57 | 602c3dfadacc7bef0998dbf0b130578ff579e32a | diff --git a/.changeset/node18-context7-mcp.md b/.changeset/node18-context7-mcp.md
new file mode 100644
index 000000000..3b2940d2d
--- /dev/null
+++ b/.changeset/node18-context7-mcp.md
@@ -0,0 +1,5 @@
+---
+"@upstash/context7-mcp": patch
+---
+
+Restore Node 18 support by pinning undici to ^6.26.0 and commander to ^13.... | [
".changeset/node18-context7-mcp.md",
"packages/mcp/package.json",
"pnpm-lock.yaml"
] | [] | true | |
upstash/context7 | 2,757 | issue_to_patch | feat(docs): add VS Code extension documentation and update all clients reference | e35cb61068fad12cafe011a6487a7a2092f66c1a | d678a35b880571d2a530f81bc3fdcf1220cf8d47 | diff --git a/docs/clients/vscode.mdx b/docs/clients/vscode.mdx
new file mode 100644
index 00000000..387e7d63
--- /dev/null
+++ b/docs/clients/vscode.mdx
@@ -0,0 +1,76 @@
+---
+title: VS Code
+description: Using Context7 with VS Code
+---
+
+The [Context7 VS Code extension](https://marketplace.visualstudio.com/items?ite... | [
"docs/clients/vscode.mdx",
"docs/docs.json",
"docs/resources/all-clients.mdx"
] | [] | true | |||
upstash/context7 | 2,752 | issue_to_patch | feat(copilot-plugin): pass CONTEXT7_API_KEY from environment to MCP server | The Copilot CLI plugin connected to mcp.context7.com anonymously with no way to attach an API key. Add a CONTEXT7_API_KEY header sourced from the environment so plugin users are billed against their own plan.
Uses the ${VAR:-} fallback form: verified on copilot 1.0.61 that plain ${VAR} passes the literal placeholder t... | 8bcad1372fb211387ae4187d42dc11624333d9b9 | 2fdfa35e5a423640e46b7ad16dcde7cfc51fd699 | diff --git a/.github/plugin/marketplace.json b/.github/plugin/marketplace.json
index 854b3179..7342e159 100644
--- a/.github/plugin/marketplace.json
+++ b/.github/plugin/marketplace.json
@@ -12,7 +12,7 @@
"name": "context7",
"source": "./plugins/copilot/context7",
"description": "Up-to-date documen... | [
".github/plugin/marketplace.json",
"plugins/copilot/context7/.mcp.json",
"plugins/copilot/context7/README.md",
"plugins/copilot/context7/plugin.json"
] | [] | true | ||
upstash/context7 | 2,751 | issue_to_patch | update test library for redirection | 75361160c7deac6c4b003161b408268e5a875af1 | 7675622e19515e69cd78546f7bc0e38262a0af6c | diff --git a/packages/sdk/src/client.ts b/packages/sdk/src/client.ts
index 39a0f439f..e5c766549 100644
--- a/packages/sdk/src/client.ts
+++ b/packages/sdk/src/client.ts
@@ -117,7 +117,7 @@ export class Context7 {
/**
* Get documentation context for a library
* @param query The user's question or task
- * @... | [
"packages/sdk/src/client.test.ts",
"packages/sdk/src/client.ts",
"packages/sdk/src/commands/get-context/index.test.ts",
"packages/sdk/src/commands/types.ts"
] | [] | diff --git a/packages/sdk/src/client.test.ts b/packages/sdk/src/client.test.ts
index 24f9c50f1..d17ddfd53 100644
--- a/packages/sdk/src/client.test.ts
+++ b/packages/sdk/src/client.test.ts
@@ -77,7 +77,7 @@ describe("Context7 Client", () => {
const client = new Context7({ apiKey });
test("should get context... | true | ||
upstash/context7 | 2,749 | issue_to_patch | [Bug]: plugin-based install is not documented and seems like doesn't support the API keys
### MCP Client
Claude Code
### Context7 MCP Version
1.0.0 (as plugin)
### Bug Description
First of all, there is no documentation about using upstash7 [Claude Code plugins](https://github.com/upstash/context7/tree/master/plu... | feat: update plugin version to 1.0.2 and add API key usage instructions in documentation | fixes #1713 | 43e410645c7807cdc2a91a7453db102ddf442aff | c76b6b95e8dac5ac525f3fc46b107dccbb44480f | diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
index 933b6f60..9a32fe78 100644
--- a/.claude-plugin/marketplace.json
+++ b/.claude-plugin/marketplace.json
@@ -8,7 +8,7 @@
"name": "context7",
"source": "./plugins/claude/context7",
"description": "Up-to-date documentat... | [
".claude-plugin/marketplace.json",
"docs/clients/claude-code.mdx",
"plugins/claude/context7/.mcp.json",
"plugins/claude/context7/README.md"
] | [] | true | |
upstash/context7 | 2,731 | issue_to_patch | chore(release): version packages | This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to master, this PR wil... | c921c8becf356425973aa1a90e4ab94bca6593fc | 37579d8f27884fb7792ff16b92be8449f6c138b6 | diff --git a/.changeset/bump-cli-runtime-deps.md b/.changeset/bump-cli-runtime-deps.md
deleted file mode 100644
index 9414c9f1b..000000000
--- a/.changeset/bump-cli-runtime-deps.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"ctx7": patch
----
-
-Bump runtime dependencies: `commander` 13 -> 15 and `ora` 9.0 -> 9.4.
diff --git ... | [
".changeset/bump-cli-runtime-deps.md",
".changeset/bump-mcp-runtime-deps.md",
".changeset/ctx7-1604-prompts-resources-handlers.md",
".changeset/git-bash-library-id.md",
".changeset/mcp-auth-elicit.md",
".changeset/sdk-non-json-error-response.md",
".changeset/xdg-context7-dirs.md",
"packages/cli/CHANGE... | [] | true | ||
upstash/context7 | 2,716 | issue_to_patch | feat(mcp): switch anonymous sign-in nudge to elicitation | ## Summary
Replaces the in-tool-result sign-in nudge introduced in #2604 with an MCP `elicitation/create` form request. The previous markdown block instructed the assistant to relay the message to the user and offer to run the setup command, which some agents (notably Claude Code) flagged as prompt injection on the to... | c436700959bf0c277aa02c57195b5a24778cf8aa | a339e1d1bc675a411224a4de43d0c3687f1aae64 | diff --git a/.changeset/mcp-auth-elicit.md b/.changeset/mcp-auth-elicit.md
new file mode 100644
index 000000000..31f59541a
--- /dev/null
+++ b/.changeset/mcp-auth-elicit.md
@@ -0,0 +1,11 @@
+---
+"@upstash/context7-mcp": minor
+---
+
+Replace the in-result sign-in nudge with an MCP form elicitation. When the backend si... | [
".changeset/mcp-auth-elicit.md",
"packages/mcp/src/index.ts",
"packages/mcp/src/lib/auth/auth-prompt.ts"
] | [] | true | ||
upstash/context7 | 2,734 | issue_to_patch | Docker MCP Toolkit
I am trying to use the `context7` server with the Docker MCP Toolkit, all other MCP servers I am using are working well, but this one is giving me errors when trying to connect via VS Code. I did a `docker pull mcp/context7` and added the server to my settings file:
```
// other servers above
"cont... | docs: document Docker MCP Toolkit stdio transport | Fixes #300
## Summary
- Add a README note for the Docker MCP Toolkit `mcp/context7` image.
- Document setting `MCP_TRANSPORT=stdio` for stdio-based clients such as VS Code, Cline, Roo Code, and Claude Desktop.
- Leave the Dockerfile, server defaults, and runtime behavior unchanged so HTTP and Smithery-oriented flows a... | 428af3e06753d92d8ec5879de097868232d8bef3 | 96254c6309f16d889ec3d3e4d104acaa74a68932 | diff --git a/docs/resources/all-clients.mdx b/docs/resources/all-clients.mdx
index 0df2e9ea5..3177f8808 100644
--- a/docs/resources/all-clients.mdx
+++ b/docs/resources/all-clients.mdx
@@ -959,7 +959,7 @@ qwen mcp add context7 npx -y @upstash/context7-mcp --api-key YOUR_API_KEY
FROM node:18-alpine
WORKDIR /app
RUN n... | [
"docs/resources/all-clients.mdx",
"packages/mcp/README.md"
] | [] | true | |
upstash/context7 | 2,748 | issue_to_patch | feat: add Context7 Codex plugin with installation instructions and documentation | 1f6212baa889c8e3685db7c740b173e61f8702cb | 402271e354f792f688d083b88b9099228c87a93e | diff --git a/.agents/plugins/marketplace.json b/.agents/plugins/marketplace.json
new file mode 100644
index 000000000..a017b9b9c
--- /dev/null
+++ b/.agents/plugins/marketplace.json
@@ -0,0 +1,20 @@
+{
+ "name": "context7-marketplace",
+ "interface": {
+ "displayName": "Context7 Marketplace"
+ },
+ "plugins": [
... | [
".agents/plugins/marketplace.json",
".gitignore",
"docs/clients/codex.mdx",
"docs/resources/all-clients.mdx",
"plugins/codex/context7/.codex-plugin/plugin.json",
"plugins/codex/context7/.mcp.json",
"plugins/codex/context7/README.md",
"plugins/codex/context7/skills/context7-mcp/SKILL.md"
] | [] | true | |||
upstash/context7 | 2,747 | issue_to_patch | trigger the Test workflow only when they touch packages/** or the root files that affect the toolchain | cc011a470743a38839a9fd54ff160a26d7680240 | e2ca7271e154ec632e0dc1833326dbf2251d27ae | diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 9afa6a9c..19680955 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -2,6 +2,15 @@ name: Test
on:
pull_request:
+ paths:
+ - "packages/**"
+ - "package.json"
+ - "pnpm-lock.yaml"
+ - "pnpm-w... | [
".github/workflows/test.yml"
] | [] | true | |||
upstash/context7 | 2,743 | issue_to_patch | fix copilot cli plugin compatability | cc011a470743a38839a9fd54ff160a26d7680240 | 435bf26a6328823fe3450d47a8bf0b897e133b24 | diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
index d63356876..933b6f601 100644
--- a/.claude-plugin/marketplace.json
+++ b/.claude-plugin/marketplace.json
@@ -5,10 +5,10 @@
},
"plugins": [
{
- "name": "context7-plugin",
+ "name": "context7",
"source": "./plu... | [
".claude-plugin/marketplace.json",
".github/plugin/marketplace.json",
".gitignore",
"docs/clients/claude-code.mdx",
"docs/clients/copilot-cli.mdx",
"docs/docs.json",
"docs/resources/all-clients.mdx",
"plugins/claude/context7/.claude-plugin/plugin.json",
"plugins/claude/context7/.mcp.json",
"plugin... | [] | true | |||
upstash/context7 | 2,742 | issue_to_patch | feat: add Codex client documentation and update OAuth flow details for existing clients | 0807914d91de553d0803b945b673022b6c2a03e4 | 2e18e20068fae1f74c9cb86724c61a5f6e701a20 | diff --git a/docs/clients/claude-code.mdx b/docs/clients/claude-code.mdx
index 75995453..f8289582 100644
--- a/docs/clients/claude-code.mdx
+++ b/docs/clients/claude-code.mdx
@@ -13,17 +13,7 @@ Run the setup command to configure Context7 for Claude Code:
npx ctx7 setup --claude
```
-This flow authenticates via OAut... | [
"docs/clients/claude-code.mdx",
"docs/clients/cli.mdx",
"docs/clients/codex.mdx",
"docs/clients/pi.mdx",
"docs/docs.json",
"docs/resources/all-clients.mdx"
] | [] | true | |||
upstash/context7 | 2,741 | issue_to_patch | chore: remove changeset check workflow | Removes the Changeset Check job that runs on every pull request.
- Deletes `.github/workflows/changeset-check.yml`, which failed PRs lacking a `.changeset/*.md` file unless they carried the `skip-changeset` label. | 0807914d91de553d0803b945b673022b6c2a03e4 | 66990d06a81d74ee303dbe9fd979b6adcb5675ab | diff --git a/.github/workflows/changeset-check.yml b/.github/workflows/changeset-check.yml
deleted file mode 100644
index c4fccc6b..00000000
--- a/.github/workflows/changeset-check.yml
+++ /dev/null
@@ -1,47 +0,0 @@
-name: Changeset Check
-
-on:
- pull_request:
- types: [opened, synchronize, reopened, labeled, unla... | [
".github/workflows/changeset-check.yml"
] | [] | true |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.