scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-14 23:21:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 102

fix(friendli): emit declared reasoning-effort levels and harden weekly sync

Browse source 
The transform now declares reasoning_effort_levels straight from the catalog's
effort options instead of mapping them onto per-level support flags, which
mis-advertised efforts these models do not take. max_tokens mirrors
max_completion_tokens rather than context_length, supports_prompt_caching is
derived from cache-read pricing, video input is read from input_modalities, and
a failed Friendli fetch no longer breaks the weekly sync run. The committed
friendliai entries are regenerated from the live catalog: stale gemma-4 token
caps refreshed, the delisted K-EXAONE-236B-A23B entry dropped, and GLM-5.3 /
GLM-5.3-Flash picked up with current billed pricing.
This commit is contained in:
mateo-berri 2026-08-31 13:34:48 -07:00
parent b346dd414b
commit b4a9ddb924
3 changed files with 331 additions and 158 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

57
.github/scripts/auto_update_price_and_context_window_file.py vendored
View file

@ -27,35 +27,13 @@ FRIENDLI_API_URL = "https://api.friendli.ai/serverless/v1/models"
FRIENDLI_PROVIDER = "friendliai"
INHERITABLE_BASE_KEYS = (
"supports_reasoning",
"supports_function_calling",
"supports_parallel_function_calling",
"supports_response_schema",
"supports_system_messages",
"supports_tool_choice",
"supports_vision",
"supports_pdf_input",
"supports_prompt_caching",
"supports_assistant_prefill",
"supports_low_reasoning_effort",
"supports_minimal_reasoning_effort",
"supports_max_reasoning_effort",
"supports_xhigh_reasoning_effort",
"supports_none_reasoning_effort",
"supports_adaptive_thinking",
"supports_output_config",
"supports_native_structured_output",
)
EFFORT_FLAG_MAP = {
"none": "supports_none_reasoning_effort",
"minimal": "supports_minimal_reasoning_effort",
"low": "supports_low_reasoning_effort",
"medium": "supports_low_reasoning_effort",
"high": "supports_max_reasoning_effort",
"xhigh": "supports_xhigh_reasoning_effort",
"max": "supports_max_reasoning_effort",
}
REASONING_EFFORT_LEVEL_ORDER = ("none", "minimal", "low", "medium", "high", "xhigh", "max")
def _find_base_model_entry(base_model: str, local_data: dict) -> str | None:
@ -72,15 +50,14 @@ def _find_base_model_entry(base_model: str, local_data: dict) -> str | None:
return None
def _effort_flags(reasoning_options: list) -> dict:
flags: dict[str, bool] = {flag: False for flag in EFFORT_FLAG_MAP.values()}
for opt in reasoning_options or []:
if opt.get("type") == "effort":
for val in opt.get("values", []):
flag = EFFORT_FLAG_MAP.get(val)
if flag:
flags[flag] = True
return flags
def _reasoning_effort_levels(reasoning_options: list) -> list:
offered = {
val
for opt in reasoning_options or []
if opt.get("type") == "effort"
for val in opt.get("values", [])
}
return [level for level in REASONING_EFFORT_LEVEL_ORDER if level in offered]
def _pricing(pricing: dict) -> dict:
@ -97,15 +74,19 @@ def _pricing(pricing: dict) -> dict:
def _modality_flags(input_mods: list) -> dict:
has_image = "image" in (input_mods or [])
mods = input_mods or []
has_image = "image" in mods
return {
"supports_vision": has_image,
"supports_image_input": has_image,
"supports_video_input": "video" in mods,
}
def transform_friendli_data(data: list, local_data: dict) -> dict:
transformed: dict[str, dict] = {}
if not data:
return transformed
for model in data:
model_id = model["id"]
base_model = model.get("base_model") or ""
@ -123,17 +104,21 @@ def transform_friendli_data(data: list, local_data: dict) -> dict:
ctx = model.get("context_length")
if ctx is not None:
entry["max_input_tokens"] = int(ctx)
entry["max_tokens"] = int(ctx)
max_out = model.get("max_completion_tokens")
if max_out is not None:
entry["max_output_tokens"] = int(max_out)
entry["max_tokens"] = int(max_out)
entry.update(_pricing(model.get("pricing", {})))
pricing = _pricing(model.get("pricing", {}))
entry.update(pricing)
entry["supports_prompt_caching"] = "cache_read_input_token_cost" in pricing
reasoning = model.get("reasoning") is True
entry["supports_reasoning"] = reasoning
if reasoning:
entry.update(_effort_flags(model.get("reasoning_options", [])))
entry["reasoning_effort_levels"] = _reasoning_effort_levels(
model.get("reasoning_options", [])
)
func = model.get("functionality", {})
entry["supports_function_calling"] = func.get("tool_call") is True

296
model_prices_and_context_window.json
View file

@ -19564,145 +19564,49 @@
"supports_system_messages": true,
"supports_tool_choice": true
},
"friendliai/LGAI-EXAONE/K-EXAONE-236B-A23B": {
"friendliai/zai-org/GLM-5.3-Flash": {
"litellm_provider": "friendliai",
"max_input_tokens": 262144,
"max_tokens": 262144,
"max_output_tokens": 262144,
"input_cost_per_token": 2e-07,
"output_cost_per_token": 8e-07,
"cache_read_input_token_cost": 1e-07,
"supports_reasoning": true,
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"mode": "chat",
"comment": "Open multilingual MoE model for reasoning, agentic tool use, and long-context work with strong Korean capabilities",
"deprecation_date": "2026-08-20",
"source": "https://api.friendli.ai/serverless/v1/models",
"supports_vision": false,
"supports_image_input": false
},
"friendliai/MiniMaxAI/MiniMax-M2.5": {
"litellm_provider": "friendliai",
"max_input_tokens": 196608,
"max_tokens": 196608,
"max_output_tokens": 196608,
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"cache_read_input_token_cost": 6e-08,
"supports_reasoning": true,
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"mode": "chat",
"comment": "Prior MiniMax coding model for agent workflows, office edits, and automation",
"source": "https://api.friendli.ai/serverless/v1/models",
"supports_vision": false,
"supports_image_input": false
},
"friendliai/deepseek-ai/DeepSeek-V3.2": {
"litellm_provider": "friendliai",
"supports_reasoning": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"max_input_tokens": 1048576,
"max_output_tokens": 1048576,
"max_tokens": 1048576,
"input_cost_per_token": 1.5e-07,
"output_cost_per_token": 5e-07,
"cache_read_input_token_cost": 3e-08,
"supports_prompt_caching": true,
"supports_assistant_prefill": true,
"max_input_tokens": 163840,
"max_tokens": 163840,
"max_output_tokens": 163840,
"input_cost_per_token": 5e-07,
"output_cost_per_token": 1.5e-06,
"cache_read_input_token_cost": 2.5e-07,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"mode": "chat",
"comment": "DeepSeek chat model for instruction following, coding, and analysis",
"source": "https://api.friendli.ai/serverless/v1/models",
"supports_vision": false,
"supports_image_input": false
},
"friendliai/zai-org/GLM-5.1": {
"litellm_provider": "friendliai",
"supports_reasoning": true,
"reasoning_effort_levels": [
"low",
"high",
"max"
],
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_prompt_caching": true,
"max_input_tokens": 202752,
"max_tokens": 202752,
"max_output_tokens": 202752,
"input_cost_per_token": 1.4e-06,
"output_cost_per_token": 4.4e-06,
"cache_read_input_token_cost": 2.6e-07,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"mode": "chat",
"comment": "Strong GLM coding model for agentic engineering, terminals, and repository generation",
"source": "https://api.friendli.ai/serverless/v1/models",
"supports_vision": false,
"supports_image_input": false
},
"friendliai/google/gemma-4-31B-it": {
"litellm_provider": "friendliai",
"supports_tool_choice": true,
"supports_vision": true,
"max_input_tokens": 8192,
"max_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 1.4e-07,
"output_cost_per_token": 4e-07,
"supports_reasoning": true,
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_image_input": true,
"supports_video_input": true,
"mode": "chat",
"comment": "Largest Gemma 4 instruction model for open, self-hosted chat and reasoning",
"comment": "Native multimodal GLM model for efficient coding and long-horizon agent tasks",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"friendliai/zai-org/GLM-5.2": {
"friendliai/zai-org/GLM-5.3": {
"litellm_provider": "friendliai",
"supports_reasoning": true,
"supports_function_calling": true,
"max_input_tokens": 1048576,
"max_tokens": 1048576,
"max_output_tokens": 1048576,
"input_cost_per_token": 1.4e-06,
"output_cost_per_token": 4.4e-06,
"cache_read_input_token_cost": 2.6e-07,
"supports_max_reasoning_effort": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"mode": "chat",
"comment": "Open flagship GLM for long-horizon coding agents and million-token context work",
"source": "https://api.friendli.ai/serverless/v1/models",
"supports_vision": false,
"supports_image_input": false
},
"friendliai/LGAI-EXAONE/K-EXAONE-2.0-750B-A37B": {
"litellm_provider": "friendliai",
"max_input_tokens": 262144,
"max_tokens": 262144,
"max_output_tokens": 262144,
"input_cost_per_token": 6e-07,
"output_cost_per_token": 2.4e-06,
"cache_read_input_token_cost": 1.2e-07,
"max_tokens": 1048576,
"input_cost_per_token": 1.26e-06,
"output_cost_per_token": 3.96e-06,
"cache_read_input_token_cost": 2.34e-07,
"supports_prompt_caching": true,
"supports_reasoning": true,
"reasoning_effort_levels": [
"low",
"high",
"max"
],
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
@ -19711,8 +19615,156 @@
"supports_tool_choice": true,
"supports_vision": false,
"supports_image_input": false,
"supports_video_input": false,
"mode": "chat",
"comment": "Flagship GLM model for long-horizon coding, agents, and complex project delivery",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"friendliai/google/gemma-4-31B-it": {
"litellm_provider": "friendliai",
"max_input_tokens": 262144,
"max_output_tokens": 262144,
"max_tokens": 262144,
"input_cost_per_token": 1.4e-07,
"output_cost_per_token": 4e-07,
"supports_prompt_caching": false,
"supports_reasoning": true,
"reasoning_effort_levels": [],
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_vision": true,
"supports_image_input": true,
"supports_video_input": false,
"mode": "chat",
"comment": "Largest Gemma 4 instruction model for open, self-hosted chat and reasoning",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"friendliai/zai-org/GLM-5.2": {
"litellm_provider": "friendliai",
"max_input_tokens": 1048576,
"max_output_tokens": 1048576,
"max_tokens": 1048576,
"input_cost_per_token": 1.4e-06,
"output_cost_per_token": 4.4e-06,
"cache_read_input_token_cost": 2.6e-07,
"supports_prompt_caching": true,
"supports_reasoning": true,
"reasoning_effort_levels": [
"high",
"max"
],
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_vision": false,
"supports_image_input": false,
"supports_video_input": false,
"mode": "chat",
"comment": "Open flagship GLM for long-horizon coding agents and million-token context work",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"friendliai/LGAI-EXAONE/K-EXAONE-2.0-750B-A37B": {
"litellm_provider": "friendliai",
"max_input_tokens": 262144,
"max_output_tokens": 262144,
"max_tokens": 262144,
"input_cost_per_token": 6e-07,
"output_cost_per_token": 2.4e-06,
"cache_read_input_token_cost": 1.2e-07,
"supports_prompt_caching": true,
"supports_reasoning": true,
"reasoning_effort_levels": [],
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_vision": false,
"supports_image_input": false,
"supports_video_input": false,
"mode": "chat",
"comment": "Frontier-scale multilingual language model developed by LG AI Research",
"deprecation_date": "2026-09-06",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"friendliai/deepseek-ai/DeepSeek-V3.2": {
"litellm_provider": "friendliai",
"max_input_tokens": 163840,
"max_output_tokens": 163840,
"max_tokens": 163840,
"input_cost_per_token": 5e-07,
"output_cost_per_token": 1.5e-06,
"cache_read_input_token_cost": 2.5e-07,
"supports_prompt_caching": true,
"supports_reasoning": true,
"reasoning_effort_levels": [],
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_vision": false,
"supports_image_input": false,
"supports_video_input": false,
"mode": "chat",
"comment": "DeepSeek chat model for instruction following, coding, and analysis",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"friendliai/MiniMaxAI/MiniMax-M2.5": {
"litellm_provider": "friendliai",
"max_input_tokens": 196608,
"max_output_tokens": 196608,
"max_tokens": 196608,
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"cache_read_input_token_cost": 6e-08,
"supports_prompt_caching": true,
"supports_reasoning": true,
"reasoning_effort_levels": [],
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_vision": false,
"supports_image_input": false,
"supports_video_input": false,
"mode": "chat",
"comment": "Prior MiniMax coding model for agent workflows, office edits, and automation",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"friendliai/zai-org/GLM-5.1": {
"litellm_provider": "friendliai",
"max_input_tokens": 202752,
"max_output_tokens": 202752,
"max_tokens": 202752,
"input_cost_per_token": 1.4e-06,
"output_cost_per_token": 4.4e-06,
"cache_read_input_token_cost": 2.6e-07,
"supports_prompt_caching": true,
"supports_reasoning": true,
"reasoning_effort_levels": [],
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_response_schema": true,
"supports_native_structured_output": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"supports_vision": false,
"supports_image_input": false,
"supports_video_input": false,
"mode": "chat",
"comment": "Strong GLM coding model for agentic engineering, terminals, and repository generation",
"source": "https://api.friendli.ai/serverless/v1/models"
},
"ft:babbage-002": {

136
tests/test_litellm/test_auto_update_price_and_context_window_file.py Normal file
View file

@ -0,0 +1,136 @@
"""Unit tests for the Friendli transform in
`.github/scripts/auto_update_price_and_context_window_file.py`."""
from __future__ import annotations
import importlib.util
import sys
from pathlib import Path
import pytest
SCRIPT_PATH = (
Path(__file__).resolve().parents[2]
/ ".github"
/ "scripts"
/ "auto_update_price_and_context_window_file.py"
)
@pytest.fixture(scope="module")
def sync_module():
spec = importlib.util.spec_from_file_location(
"auto_update_price_and_context_window_file", SCRIPT_PATH
)
assert spec and spec.loader, f"Could not load spec for {SCRIPT_PATH}"
module = importlib.util.module_from_spec(spec)
sys.modules["auto_update_price_and_context_window_file"] = module
spec.loader.exec_module(module)
return module
def _reasoning_model(**overrides: object) -> dict:
model = {
"id": "zai-org/GLM-Test",
"base_model": "zhipuai/glm-test",
"context_length": 1048576,
"max_completion_tokens": 131072,
"pricing": {"input": "0.00000015", "output": "0.0000005", "input_cache_read": "0.00000003"},
"reasoning": True,
"reasoning_options": [{"type": "effort", "values": ["max", "high", "low"]}],
"functionality": {
"tool_call": True,
"parallel_tool_call": True,
"structured_output": True,
"system_messages": True,
"tool_choice": True,
},
"input_modalities": ["text", "image", "video"],
"mode": "chat",
}
model.update(overrides)
return model
def test_transform_emits_declared_effort_levels_in_canonical_order(sync_module):
entry = sync_module.transform_friendli_data([_reasoning_model()], {})[
"friendliai/zai-org/GLM-Test"
]
assert entry["supports_reasoning"] is True
assert entry["reasoning_effort_levels"] == ["low", "high", "max"]
assert not any(k.endswith("_reasoning_effort") for k in entry)
def test_transform_reasoning_model_without_effort_options_declares_empty_levels(sync_module):
model = _reasoning_model(reasoning_options=[{"type": "budget_tokens", "values": []}])
entry = sync_module.transform_friendli_data([model], {})["friendliai/zai-org/GLM-Test"]
assert entry["reasoning_effort_levels"] == []
def test_transform_non_reasoning_model_declares_no_levels(sync_module):
model = _reasoning_model(reasoning=False, reasoning_options=[])
entry = sync_module.transform_friendli_data([model], {})["friendliai/zai-org/GLM-Test"]
assert entry["supports_reasoning"] is False
assert "reasoning_effort_levels" not in entry
def test_transform_max_tokens_mirrors_output_cap_not_context(sync_module):
entry = sync_module.transform_friendli_data([_reasoning_model()], {})[
"friendliai/zai-org/GLM-Test"
]
assert entry["max_input_tokens"] == 1048576
assert entry["max_output_tokens"] == 131072
assert entry["max_tokens"] == entry["max_output_tokens"]
def test_transform_prompt_caching_follows_cache_pricing(sync_module):
cached = sync_module.transform_friendli_data([_reasoning_model()], {})[
"friendliai/zai-org/GLM-Test"
]
assert cached["supports_prompt_caching"] is True
assert cached["cache_read_input_token_cost"] == 3e-08
uncached_model = _reasoning_model(pricing={"input": "0.00000014", "output": "0.0000004"})
uncached = sync_module.transform_friendli_data([uncached_model], {})[
"friendliai/zai-org/GLM-Test"
]
assert uncached["supports_prompt_caching"] is False
assert "cache_read_input_token_cost" not in uncached
def test_transform_modalities_set_vision_image_and_video_flags(sync_module):
entry = sync_module.transform_friendli_data([_reasoning_model()], {})[
"friendliai/zai-org/GLM-Test"
]
assert entry["supports_vision"] is True
assert entry["supports_image_input"] is True
assert entry["supports_video_input"] is True
text_only = _reasoning_model(input_modalities=["text"])
entry_text = sync_module.transform_friendli_data([text_only], {})[
"friendliai/zai-org/GLM-Test"
]
assert entry_text["supports_vision"] is False
assert entry_text["supports_image_input"] is False
assert entry_text["supports_video_input"] is False
def test_transform_survives_failed_fetch(sync_module):
assert sync_module.transform_friendli_data(None, {}) == {}
assert sync_module.transform_friendli_data([], {}) == {}
def test_transform_inherits_allowlisted_keys_from_base_model_entry(sync_module):
local = {
"zhipuai/glm-test": {
"supports_pdf_input": True,
"supports_assistant_prefill": True,
"input_cost_per_token": 9e-06,
}
}
entry = sync_module.transform_friendli_data([_reasoning_model()], local)[
"friendliai/zai-org/GLM-Test"
]
assert entry["supports_pdf_input"] is True
assert entry["supports_assistant_prefill"] is True
assert entry["input_cost_per_token"] == 1.5e-07