c22f6e48ff
Second 5-Opus reviewer round. Applying high-confidence fixes; speculative
items (gpt-5.5-pro effort restriction, o3 image_generation gating,
o-series parallel_tool_calls per-model, gpt-5.x new model prefixes,
Anthropic fast-mode + Priority exclusion UI gate, Gemini service_tier,
Kimi k2.5 toggleable thinking) deferred to follow-up because they need
type-system changes, more verification, or backend wire work.
OpenAI max-output caps — replace the 3-line table with one driven by
direct dev.openai.com per-model fetches (cross-checked against the Azure
Foundry reasoning table):
- gpt-5.4 / gpt-5.4-pro / gpt-5.4-mini / gpt-5.4-nano: 65536 -> 128000
(https://developers.openai.com/api/docs/models/gpt-5.4 "128,000 max
output tokens"; Azure table same).
- gpt-5.3-codex: 16384 -> 128000
(https://developers.openai.com/api/docs/models/gpt-5.3-codex).
- gpt-5 / gpt-5.1 / gpt-5.2: 32k default -> 128000
(https://developers.openai.com/api/docs/models/gpt-5.2 confirms
128k; Azure table extends to gpt-5/5.1).
- gpt-5.3-chat-latest and gpt-5.1-chat keep 16384 (chat-class
variants per Azure context table row).
- o1 / o3 / o3-mini / o3-pro / o4-mini / codex-mini: 32k default ->
100000 (https://developers.openai.com/api/docs/models/o3 "100,000
max output tokens"; Azure o-series table same).
Implementation: list the two 16k chat-latest ids first so the broader
`gpt-5` 128k entry doesn't shadow them.
OpenAI reasoning_effort levels:
- gpt-5.3-codex: drop "none" from levels + flip supportsOff to false.
Dev page lists the enum as low/medium/high/xhigh only — `none` is
not in the codex variant.
- o-series bucket: change prefix from ["o3"] to
["o1","o3","o4","codex-mini"]. Previously o1 / o4-mini / codex-mini
fell into NO_REASONING_CAPS so the panel HID the effort slider for
them — real UX regression for users on those ids. Azure o-series
table confirms all four accept low/medium/high reasoning_effort.
DeepSeek default_models:
- Add deepseek-v4-pro + deepseek-v4-flash alongside the legacy
deepseek-chat / deepseek-reasoner aliases. The latter retire on
2026-07-24 per https://api-docs.deepseek.com/updates; surfacing
both lets the picker keep working on cutover.
Local backend bucket split (Ollama-stricter):
- Splits the round-1 VLLM_OLLAMA_CAPABILITIES into a vLLM-specific
bucket (keeps top_k / min_p / repetition_penalty / seed on; vLLM's
SamplingParams supports all four) and an Ollama-specific bucket
that ALSO hides top_k / min_p / repetition_penalty. Ollama's OAI
translator (ollama/openai/openai.go FromChatRequest) only copies
the documented OpenAI subset on the /v1/chat/completions path that
Studio uses; the three knobs are silently dropped even though
native /api/chat would forward them via `options`. Hiding them is
the smaller fix vs adding a backend /api/chat rewrite path.
Reviewer claims verified wrong, skipped:
- _ANTHROPIC_NEW_CODE_EXEC_PREFIXES already lists opus-4-7, opus-4-6,
sonnet-4-6 (external_provider.py:337-339). No-op.
- Mistral `seed` already renamed to `random_seed` by backend at
external_provider.py:772. No-op.
- OpenRouter `isOpenRouterMandatoryReasoningModel` uses `Set.has()`
exact match, not prefix match, so deepseek/deepseek-r1-distill-*
cannot accidentally hit the always-on guard. No-op.
Tests: 63/63 sampling_params_routing tests pass; frontend tsc clean.
377 lines
15 KiB
Python
377 lines
15 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-only
|
|
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
|
|
|
|
"""
|
|
Static registry of supported external LLM providers.
|
|
|
|
All providers expose OpenAI-compatible /v1/chat/completions endpoints
|
|
with Bearer token authentication and SSE streaming support.
|
|
"""
|
|
|
|
import re
|
|
from typing import Any
|
|
|
|
PROVIDER_REGISTRY: dict[str, dict[str, Any]] = {
|
|
"openai": {
|
|
"display_name": "OpenAI",
|
|
"base_url": "https://api.openai.com/v1",
|
|
"default_models": [
|
|
"gpt-5.5",
|
|
"gpt-5.4",
|
|
"gpt-5.4-mini",
|
|
"o3",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
# Keep the model picker scoped to the current generation. The remote
|
|
# /v1/models listing returns dozens of historical snapshots, fine-tunes
|
|
# and non-chat models (embeddings, TTS, image, moderation) that we
|
|
# never want to surface in the chat UI. Filtering here so backend
|
|
# is the single source of truth.
|
|
"model_id_allowlist": re.compile(r"^(gpt-5\.[345]|gpt-4\.5|o3)(?:[-.]|$)"),
|
|
# Hide dated snapshots and the retired plain gpt-5.3 id.
|
|
"model_id_denylist": re.compile(r"^(gpt-5\.3)$|-\d{4}-\d{2}-\d{2}$"),
|
|
},
|
|
"anthropic": {
|
|
"display_name": "Anthropic",
|
|
"base_url": "https://api.anthropic.com/v1",
|
|
"default_models": [
|
|
"claude-opus-4-7",
|
|
"claude-opus-4-6",
|
|
"claude-sonnet-4-6",
|
|
"claude-opus-4-5",
|
|
"claude-sonnet-4-5",
|
|
"claude-haiku-4-5",
|
|
],
|
|
# Anthropic /v1/models returns dated snapshot ids alongside the
|
|
# canonical names (e.g. claude-3-5-sonnet-20241022). Hide the
|
|
# YYYYMMDD-suffixed variants from the picker — same intent as the
|
|
# OpenAI denylist, just a different date format (no dashes between
|
|
# year/month/day).
|
|
"model_id_denylist": re.compile(r"-\d{8}$"),
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": False,
|
|
"auth_header": "x-api-key",
|
|
"auth_prefix": "",
|
|
"extra_headers": {
|
|
"anthropic-version": "2023-06-01",
|
|
},
|
|
"openai_compatible": False,
|
|
"notes": "Native Anthropic Messages API. Uses x-api-key header and /v1/messages endpoint with SSE translation.",
|
|
},
|
|
"gemini": {
|
|
"display_name": "Google Gemini",
|
|
"base_url": "https://generativelanguage.googleapis.com/v1beta/openai",
|
|
# Curated lineup — Google's /v1beta/openai/models returns dozens
|
|
# of historical / experimental / embedding ids. Cap to the current
|
|
# 3.x family plus the rolling `*-latest` aliases.
|
|
"default_models": [
|
|
"gemini-3.1-pro-preview",
|
|
"gemini-3.1-flash-lite",
|
|
"gemini-3-flash-preview",
|
|
"gemini-pro-latest",
|
|
"gemini-flash-latest",
|
|
"gemini-flash-lite-latest",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"notes": "OpenAI-compatible endpoint. API key from https://aistudio.google.com/apikey.",
|
|
"model_id_allowlist": re.compile(
|
|
r"^(gemini-3\.1-flash-lite|gemini-3-flash-preview|"
|
|
r"gemini-3\.1-pro-preview|gemini-pro-latest|"
|
|
r"gemini-flash-latest|gemini-flash-lite-latest)$"
|
|
),
|
|
# Gemini's OpenAI-compatible layer inherits OpenAI's 4-stop cap
|
|
# (https://ai.google.dev/gemini-api/docs/openai). Without the
|
|
# explicit cap the default 16 leaks through and the upstream
|
|
# silently drops the overflow.
|
|
"stop_max": 4,
|
|
},
|
|
"deepseek": {
|
|
"display_name": "DeepSeek",
|
|
"base_url": "https://api.deepseek.com/v1",
|
|
# Legacy aliases (deepseek-chat / deepseek-reasoner) retire
|
|
# 2026-07-24 per https://api-docs.deepseek.com/updates. Surface
|
|
# the new canonical ids (deepseek-v4-flash / deepseek-v4-pro)
|
|
# alongside so the picker keeps working on cutover.
|
|
"default_models": [
|
|
"deepseek-v4-pro",
|
|
"deepseek-v4-flash",
|
|
"deepseek-chat",
|
|
"deepseek-reasoner",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": False,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"notes": "OpenAI-compatible API. deepseek-v4-pro / deepseek-v4-flash are the new canonical ids; deepseek-chat / deepseek-reasoner remain as legacy aliases until 2026-07-24.",
|
|
},
|
|
"mistral": {
|
|
"display_name": "Mistral AI",
|
|
"base_url": "https://api.mistral.ai/v1",
|
|
"default_models": [
|
|
"codestral-latest",
|
|
"devstral-latest",
|
|
"devstral-medium-latest",
|
|
"magistral-medium-latest",
|
|
"ministral-14b-latest",
|
|
"ministral-3b-latest",
|
|
"ministral-8b-latest",
|
|
"mistral-large-latest",
|
|
"mistral-medium-latest",
|
|
"mistral-small-latest",
|
|
"mistral-tiny-latest",
|
|
"mistral-vibe-cli-latest",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"model_id_allowlist": re.compile(
|
|
r"^(codestral-latest|devstral-latest|devstral-medium-latest|"
|
|
r"magistral-medium-latest|ministral-(?:14b|3b|8b)-latest|"
|
|
r"mistral-(?:large|medium|small|tiny)-latest|"
|
|
r"mistral-vibe-cli-latest)$"
|
|
),
|
|
# Mistral renames OpenAI's `seed` to `random_seed` on
|
|
# /v1/chat/completions. https://docs.mistral.ai/api/endpoint/chat
|
|
"seed_field": "random_seed",
|
|
},
|
|
"kimi": {
|
|
"display_name": "Kimi",
|
|
"base_url": "https://api.moonshot.ai/v1",
|
|
# Current Kimi model lineup per the official docs:
|
|
# https://platform.kimi.ai/docs/models
|
|
# Listing/overview endpoints used to enumerate them:
|
|
# https://platform.kimi.ai/docs/api/list-models
|
|
# https://platform.kimi.ai/docs/api/overview
|
|
# kimi-k2.6 and kimi-k2.5 are the two SoTA multimodal models we
|
|
# surface in the picker; everything else (moonshot-v1-*, dated
|
|
# k2 previews) is filtered out by model_id_allowlist below.
|
|
"default_models": [
|
|
"kimi-k2.6",
|
|
"kimi-k2.5",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"notes": "Moonshot API key. China: use base URL https://api.moonshot.cn/v1",
|
|
"model_id_allowlist": re.compile(r"^kimi-k2\.[56]$"),
|
|
# Both k2.6 and k2.5 are reasoning-class. The API rejects
|
|
# custom sampling ("invalid temperature: only 1 is allowed for
|
|
# this model", same for top_p). frequency_penalty follows the
|
|
# same lock on those models. seed and parallel_tool_calls are
|
|
# not in Kimi's documented chat schema; strip them too so a
|
|
# stale client or direct API caller cannot smuggle them onto
|
|
# the wire and 400 the request.
|
|
"body_omit": (
|
|
"temperature",
|
|
"top_p",
|
|
"frequency_penalty",
|
|
"seed",
|
|
"parallel_tool_calls",
|
|
),
|
|
# Kimi accepts at most 5 stop strings (each <= 32 bytes) per
|
|
# https://platform.kimi.ai/docs/api/chat
|
|
"stop_max": 5,
|
|
"stop_max_bytes": 32,
|
|
},
|
|
"qwen": {
|
|
"display_name": "Qwen",
|
|
"base_url": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
|
|
"default_models": [
|
|
"qwen-plus",
|
|
"qwen-turbo",
|
|
"qwen-max",
|
|
"qwen2.5-72b-instruct",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"notes": "DashScope API key. China mainland: override base URL to https://dashscope.aliyuncs.com/compatible-mode/v1",
|
|
},
|
|
"huggingface": {
|
|
"display_name": "Hugging Face",
|
|
"base_url": "https://router.huggingface.co/v1",
|
|
# Seed the picker with a few popular ids so something is selectable
|
|
# before the live /v1/models call resolves. The remote listing is
|
|
# the source of truth — see model_list_mode below.
|
|
"default_models": [
|
|
"openai/gpt-oss-120b",
|
|
"deepseek-ai/DeepSeek-V3",
|
|
"meta-llama/Llama-3.3-70B-Instruct",
|
|
"Qwen/Qwen2.5-72B-Instruct",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"notes": (
|
|
"HF token from huggingface.co/settings/tokens. Uses the "
|
|
"OpenAI-compatible router at /v1/chat/completions; /v1/models "
|
|
"returns the cross-provider chat catalog. See "
|
|
"https://huggingface.co/docs/inference-providers/index."
|
|
),
|
|
# /v1/models works on the HF router and returns the full chat-model
|
|
# catalog (state.org/model[:policy] ids). Switch to remote so users
|
|
# see live availability — the picker has a search box, and
|
|
# loadModels() merges defaults so default_models entries remain
|
|
# visible if the remote call fails.
|
|
"model_list_mode": "remote",
|
|
# Scope the catalog to first-party org repos we trust as primary
|
|
# sources. The HF /v1/models response is otherwise hundreds of
|
|
# ids long (community fine-tunes, mirrors, fp8 variants, etc.).
|
|
"model_id_allowlist": re.compile(
|
|
r"^(openai|deepseek-ai|google|meta-llama|Qwen|moonshotai|"
|
|
r"mistralai|zai-org)/"
|
|
),
|
|
# Cap the post-filter list. /v1/models has no server-side limit
|
|
# or popularity sort, so this is just "first N matches" — pair it
|
|
# with the default_models seed so the most useful flagship ids
|
|
# are always among the top regardless of the API's order.
|
|
"model_id_limit": 15,
|
|
},
|
|
"vllm": {
|
|
"display_name": "vLLM",
|
|
# User-supplied via provider_base_url; the route layer already falls
|
|
# back to the payload's base_url when the registry entry has none.
|
|
"base_url": "",
|
|
"default_models": [],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
# Force /v1/chat/completions in stream_chat_completion — vLLM's
|
|
# /v1/responses rebuilds messages and runs them through the loaded
|
|
# model's chat template, which 400s on strict-alternation templates
|
|
# (Gemma 3 raises "Conversation roles must alternate user/assistant
|
|
# /user/assistant/..."). The chat-completions path takes messages
|
|
# verbatim and avoids that template gauntlet.
|
|
"notes": "Self-hosted vLLM server. Always routed to /v1/chat/completions.",
|
|
# Surfaced through the frontend's CUSTOM_PROVIDER_PRESETS, not the
|
|
# /api/providers/registry dropdown — see list_available_providers.
|
|
"hidden": True,
|
|
},
|
|
"ollama": {
|
|
"display_name": "Ollama",
|
|
"base_url": "http://localhost:11434/v1",
|
|
"default_models": [],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"notes": (
|
|
"Local Ollama server. OpenAI-compatible /v1/chat/completions; "
|
|
"no API key. Surfaced via CUSTOM_PROVIDER_PRESETS in the frontend."
|
|
),
|
|
"hidden": True,
|
|
},
|
|
"llama_cpp": {
|
|
"display_name": "llama.cpp",
|
|
"base_url": "http://localhost:8080/v1",
|
|
"default_models": [],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"notes": (
|
|
"Local llama.cpp server (llama-server). OpenAI-compatible "
|
|
"/v1/chat/completions. Surfaced via CUSTOM_PROVIDER_PRESETS."
|
|
),
|
|
"hidden": True,
|
|
},
|
|
"openrouter": {
|
|
"display_name": "OpenRouter",
|
|
"base_url": "https://openrouter.ai/api/v1",
|
|
# Curated list for Studio's picker (explicitly locked, not live /models).
|
|
"default_models": [
|
|
"openrouter/free",
|
|
"openai/gpt-4o",
|
|
"anthropic/claude-sonnet-4-5",
|
|
"google/gemini-2.5-flash",
|
|
"mistralai/mistral-large-2411",
|
|
"deepseek/deepseek-r1",
|
|
"mistralai/mistral-small-3.1-24b-instruct",
|
|
"perceptron/perceptron-mk1",
|
|
"inclusionai/ring-2.6-1t:free",
|
|
"google/gemini-3.1-flash-lite",
|
|
"baidu/cobuddy:free",
|
|
"openai/gpt-chat-latest",
|
|
"x-ai/grok-4.3",
|
|
"ibm-granite/granite-4.1-8b",
|
|
"openrouter/owl-alpha",
|
|
"poolside/laguna-xs.2:free",
|
|
"~google/gemini-pro-latest",
|
|
"~moonshotai/kimi-latest",
|
|
],
|
|
"supports_streaming": True,
|
|
"supports_vision": True,
|
|
"supports_tool_calling": True,
|
|
"auth_header": "Authorization",
|
|
"auth_prefix": "Bearer ",
|
|
"extra_headers": {
|
|
"HTTP-Referer": "https://unsloth.ai",
|
|
"X-Title": "Unsloth Studio",
|
|
},
|
|
"notes": "Unified gateway to 300+ models across all major providers. HTTP-Referer and X-Title headers sent for attribution.",
|
|
"model_list_mode": "curated",
|
|
# OpenRouter normalises to OpenAI's chat schema and inherits
|
|
# the 4-entry stop cap.
|
|
"stop_max": 4,
|
|
},
|
|
}
|
|
|
|
|
|
def get_provider_info(provider_type: str) -> dict[str, Any] | None:
|
|
"""Return the registry entry for a provider type, or None if unknown."""
|
|
return PROVIDER_REGISTRY.get(provider_type)
|
|
|
|
|
|
def get_base_url(provider_type: str) -> str | None:
|
|
"""Return the default base URL for a provider type."""
|
|
info = PROVIDER_REGISTRY.get(provider_type)
|
|
return info["base_url"] if info else None
|
|
|
|
|
|
def list_available_providers() -> list[dict[str, Any]]:
|
|
"""Return all registered providers (for the /registry endpoint).
|
|
|
|
Hidden entries (``"hidden": True``) are filtered out — they exist in the
|
|
registry only for backend lookups (e.g. ``supports_vision`` for vLLM) and
|
|
are surfaced in the frontend via ``CUSTOM_PROVIDER_PRESETS`` instead of
|
|
the cloud-provider dropdown.
|
|
"""
|
|
result = []
|
|
for provider_type, info in PROVIDER_REGISTRY.items():
|
|
if info.get("hidden"):
|
|
continue
|
|
result.append(
|
|
{
|
|
"provider_type": provider_type,
|
|
"display_name": info["display_name"],
|
|
"base_url": info["base_url"],
|
|
"default_models": info["default_models"],
|
|
"supports_streaming": info["supports_streaming"],
|
|
"supports_vision": info.get("supports_vision", False),
|
|
"supports_tool_calling": info.get("supports_tool_calling", False),
|
|
"model_list_mode": info.get("model_list_mode", "remote"),
|
|
}
|
|
)
|
|
return result
|