Compare commits
No commits in common. "feat/model-capabilities-fields" and "main" have entirely different histories.
feat/model
...
main
@ -1,12 +1,5 @@
|
||||
# Changelog
|
||||
|
||||
## [Unreleased]
|
||||
|
||||
### Added
|
||||
- `LLMModel`: `context_length`, `default_max_output_tokens`, `precision` ve tipli `capabilities` alt-nesnesi (`tools`, `thinking`, `reasoning_effort`, `reasoning_effort_values`, `reasoning_effort_aliases`, `vision`, `max_images_per_request`, `response_format`, `seed`, `top_k`, `stop`) — hem `models()` hem `get_model()` yanıtında dolu; eski geçit yanıtlarında yoksa `None` kalır (geriye dönük uyumlu)
|
||||
- `get_model(model)` / `AsyncEvrenLLMClient.get_model(model)`: tek bir modelin detayı (`GET /v1/models/{model}`) — `LLMModel`; bilinmeyen model `LLMNotFoundError` fırlatır
|
||||
- `LLMModelCapabilities` dataclass'ı dışa açıldı (`from evren_sdk import LLMModelCapabilities`)
|
||||
|
||||
## [0.11.1] - 2026-11-04
|
||||
|
||||
### Fixed
|
||||
|
||||
16
README.md
16
README.md
@ -446,20 +446,12 @@ except InferenceError:
|
||||
| `speak(model, text, voice=, response_format=, instructions=)` | `/v1/audio/speech` | `SpeechResult` |
|
||||
| `respond(model, input, **kw)` / `get_response(id)` | `/v1/responses` | `ResponseResult` |
|
||||
| `models()` / `list_models()` | `/v1/models` | `list[LLMModel]` / `list[str]` |
|
||||
| `get_model(model)` | `/v1/models/{model}` | `LLMModel` |
|
||||
| `quota()` | `/v1/quota` | `QuotaStatus` |
|
||||
| `request_status(request_id)` | `/v1/requests/{id}` | `RequestStatus` |
|
||||
|
||||
`encode_image(kaynak, max_side=None)` yol, bayt ya da `data:` URL'yi geçidin kabul
|
||||
ettiği gömülü biçime çevirir.
|
||||
|
||||
`LLMModel` artık `context_length`, `default_max_output_tokens`, `precision` ve
|
||||
tipli `capabilities` (`tools`, `thinking`, `reasoning_effort`,
|
||||
`reasoning_effort_values`, `reasoning_effort_aliases`, `vision`,
|
||||
`max_images_per_request`, `response_format`, `seed`, `top_k`, `stop`) alanlarını
|
||||
taşır — bir istek göndermeden önce modelin hangi parametreyi desteklediğini
|
||||
denetlemek için. Eski geçit yanıtlarında bu alanlar yoksa `None` kalır.
|
||||
|
||||
</details>
|
||||
|
||||
---
|
||||
@ -837,20 +829,12 @@ except InferenceError:
|
||||
| `speak(model, text, voice=, response_format=, instructions=)` | `/v1/audio/speech` | `SpeechResult` |
|
||||
| `respond(model, input, **kw)` / `get_response(id)` | `/v1/responses` | `ResponseResult` |
|
||||
| `models()` / `list_models()` | `/v1/models` | `list[LLMModel]` / `list[str]` |
|
||||
| `get_model(model)` | `/v1/models/{model}` | `LLMModel` |
|
||||
| `quota()` | `/v1/quota` | `QuotaStatus` |
|
||||
| `request_status(request_id)` | `/v1/requests/{id}` | `RequestStatus` |
|
||||
|
||||
`encode_image(source, max_side=None)` turns a path, bytes or `data:` URL into the
|
||||
embedded form the gateway accepts.
|
||||
|
||||
`LLMModel` now carries `context_length`, `default_max_output_tokens`, `precision`
|
||||
and a typed `capabilities` (`tools`, `thinking`, `reasoning_effort`,
|
||||
`reasoning_effort_values`, `reasoning_effort_aliases`, `vision`,
|
||||
`max_images_per_request`, `response_format`, `seed`, `top_k`, `stop`) — check what
|
||||
a model supports before sending a request that would otherwise 400. Fields stay
|
||||
`None` when an older gateway response doesn't include them.
|
||||
|
||||
</details>
|
||||
|
||||
---
|
||||
|
||||
@ -32,7 +32,6 @@ from .llm_tasks import (
|
||||
DecisionBatchResult,
|
||||
DecisionResult,
|
||||
LLMModel,
|
||||
LLMModelCapabilities,
|
||||
OcrResult,
|
||||
QuotaStatus,
|
||||
RequestStatus,
|
||||
@ -77,7 +76,6 @@ __all__ = [
|
||||
"LLMConflictError",
|
||||
"LLMInsufficientCreditsError",
|
||||
"LLMModel",
|
||||
"LLMModelCapabilities",
|
||||
"LLMNotFoundError",
|
||||
"LLMRateLimitError",
|
||||
"LLMServiceError",
|
||||
|
||||
@ -26,7 +26,6 @@ from .llm_tasks import (
|
||||
decision_body,
|
||||
parse_decision,
|
||||
parse_decision_batch,
|
||||
parse_model,
|
||||
parse_models,
|
||||
parse_ocr,
|
||||
parse_quota,
|
||||
@ -146,14 +145,6 @@ class AsyncEvrenLLMClient:
|
||||
async def models(self) -> list[LLMModel]:
|
||||
return parse_models((await self._send("GET", "/v1/models")).json())
|
||||
|
||||
async def get_model(self, model: str) -> LLMModel:
|
||||
"""Tek bir modelin detayı (`/v1/models/{model}`) — `context_length`, `capabilities` vb. dahil.
|
||||
|
||||
Raises:
|
||||
LLMNotFoundError: `model` geçitte tanımlı değilse.
|
||||
"""
|
||||
return parse_model((await self._send("GET", f"/v1/models/{model}")).json())
|
||||
|
||||
async def decide(
|
||||
self, model: str, state: Any, questions: dict[str, Any], *, images: list[ImageSource] | None = None,
|
||||
idempotency_key: str | None = None, **kwargs: Any,
|
||||
|
||||
@ -32,7 +32,6 @@ from .llm_tasks import (
|
||||
decision_body,
|
||||
parse_decision,
|
||||
parse_decision_batch,
|
||||
parse_model,
|
||||
parse_models,
|
||||
parse_ocr,
|
||||
parse_quota,
|
||||
@ -208,14 +207,6 @@ class EvrenLLMClient:
|
||||
"""Anahtarın erişebildiği modeller, görev (`task`) ve modaliteleriyle."""
|
||||
return parse_models(self._send("GET", "/v1/models").json())
|
||||
|
||||
def get_model(self, model: str) -> LLMModel:
|
||||
"""Tek bir modelin detayı (`/v1/models/{model}`) — `context_length`, `capabilities` vb. dahil.
|
||||
|
||||
Raises:
|
||||
LLMNotFoundError: `model` geçitte tanımlı değilse.
|
||||
"""
|
||||
return parse_model(self._send("GET", f"/v1/models/{model}").json())
|
||||
|
||||
def decide(
|
||||
self, model: str, state: Any, questions: dict[str, Any], *, images: list[ImageSource] | None = None,
|
||||
idempotency_key: str | None = None, **kwargs: Any,
|
||||
|
||||
@ -16,37 +16,9 @@ from .llm_media import ImageSource, encode_image
|
||||
from .llm_types import JSON, ResponseMeta, parse_meta
|
||||
|
||||
|
||||
@dataclass(slots=True)
|
||||
class LLMModelCapabilities:
|
||||
"""`/v1/models` ve `/v1/models/{model}` yanıtlarının `capabilities` alanı.
|
||||
|
||||
Hangi istek parametresinin bu modelde kabul edildiğini söyler; `400` almadan
|
||||
önce denetlemek için kullanılır. Tüm alanlar geriye dönük uyum için `None`
|
||||
varsayılanlıdır — eski bir geçit yanıtı bu alanı hiç döndürmeyebilir.
|
||||
"""
|
||||
|
||||
tools: bool | None = None
|
||||
thinking: bool | None = None
|
||||
reasoning_effort: bool | None = None
|
||||
reasoning_effort_values: list[str] | None = None
|
||||
reasoning_effort_aliases: dict[str, str] | None = None
|
||||
vision: bool | None = None
|
||||
max_images_per_request: int | None = None
|
||||
response_format: bool | None = None
|
||||
seed: bool | None = None
|
||||
top_k: bool | None = None
|
||||
stop: bool | None = None
|
||||
|
||||
|
||||
@dataclass(slots=True)
|
||||
class LLMModel:
|
||||
"""`/v1/models` kaydı (`models()`) ya da `/v1/models/{model}` tekil kaydı (`get_model()`).
|
||||
|
||||
`task` uç seçimini belirler: decision -> `decide()`, ocr -> `ocr()` ...
|
||||
`context_length`, `default_max_output_tokens`, `capabilities`, `precision` hem
|
||||
liste hem detay ucundan gelir; eski bir geçit sürümü bunları döndürmezse `None`
|
||||
kalır (geriye dönük uyumlu).
|
||||
"""
|
||||
"""`/v1/models` kaydı. `task` uç seçimini belirler: decision -> `decide()`, ocr -> `ocr()` ..."""
|
||||
|
||||
id: str
|
||||
task: str = "chat"
|
||||
@ -54,10 +26,6 @@ class LLMModel:
|
||||
owned_by: str = "evren"
|
||||
created: int = 0
|
||||
pricing: JSON | None = None
|
||||
context_length: int | None = None
|
||||
default_max_output_tokens: int | None = None
|
||||
capabilities: LLMModelCapabilities | None = None
|
||||
precision: str | None = None
|
||||
|
||||
|
||||
@dataclass(slots=True)
|
||||
@ -225,34 +193,14 @@ def _float(value: Any) -> float | None:
|
||||
return None if value is None else float(value)
|
||||
|
||||
|
||||
def _parse_capabilities(c: JSON | None) -> LLMModelCapabilities | None:
|
||||
if c is None:
|
||||
return None
|
||||
return LLMModelCapabilities(
|
||||
tools=c.get("tools"), thinking=c.get("thinking"), reasoning_effort=c.get("reasoning_effort"),
|
||||
reasoning_effort_values=c.get("reasoning_effort_values"),
|
||||
reasoning_effort_aliases=c.get("reasoning_effort_aliases"), vision=c.get("vision"),
|
||||
max_images_per_request=c.get("max_images_per_request"), response_format=c.get("response_format"),
|
||||
seed=c.get("seed"), top_k=c.get("top_k"), stop=c.get("stop"),
|
||||
)
|
||||
|
||||
|
||||
def _parse_model(m: JSON) -> LLMModel:
|
||||
return LLMModel(
|
||||
id=str(m["id"]), task=str(m.get("task") or "chat"), modalities=list(m.get("modalities") or ["text"]),
|
||||
owned_by=str(m.get("owned_by") or "evren"), created=int(m.get("created") or 0), pricing=m.get("pricing"),
|
||||
context_length=m.get("context_length"), default_max_output_tokens=m.get("default_max_output_tokens"),
|
||||
capabilities=_parse_capabilities(m.get("capabilities")), precision=m.get("precision"),
|
||||
)
|
||||
|
||||
|
||||
def parse_models(body: JSON) -> list[LLMModel]:
|
||||
return [_parse_model(m) for m in body.get("data") or []]
|
||||
|
||||
|
||||
def parse_model(body: JSON) -> LLMModel:
|
||||
"""`GET /v1/models/{model}` kaydı — tek bir modelin görev/modalite/fiyat/kapasite bilgisi."""
|
||||
return _parse_model(body)
|
||||
return [
|
||||
LLMModel(
|
||||
id=str(m["id"]), task=str(m.get("task") or "chat"), modalities=list(m.get("modalities") or ["text"]),
|
||||
owned_by=str(m.get("owned_by") or "evren"), created=int(m.get("created") or 0), pricing=m.get("pricing"),
|
||||
)
|
||||
for m in body.get("data") or []
|
||||
]
|
||||
|
||||
|
||||
def parse_decision(body: JSON, headers: httpx.Headers | None, requested_model: str) -> DecisionResult:
|
||||
|
||||
@ -10,7 +10,6 @@ import evren_sdk.llm_client as smod
|
||||
from evren_sdk import (
|
||||
AsyncEvrenLLMClient,
|
||||
EvrenLLMClient,
|
||||
LLMNotFoundError,
|
||||
LLMRateLimitError,
|
||||
LLMValidationError,
|
||||
encode_image,
|
||||
@ -209,68 +208,6 @@ def test_models():
|
||||
models = _client(lambda req: httpx.Response(200, json=body)).models()
|
||||
assert models[0].task == "decision" and "image" in models[0].modalities
|
||||
assert models[1].task == "chat" and models[1].modalities == ["text"]
|
||||
# Eski (capability alanlarından önceki) geçit yanıtı — geriye dönük uyum, None varsayılanı
|
||||
assert models[0].context_length is None and models[0].default_max_output_tokens is None
|
||||
assert models[0].capabilities is None and models[0].precision is None
|
||||
|
||||
|
||||
# Gerçek `GET /v1/models` üzerinden canlı doğrulanmış gövde (glm-5.3, 2026-11-07)
|
||||
GLM_53_DETAIL = {
|
||||
"id": "glm-5.3", "object": "model", "created": 1789343625, "owned_by": "evren", "task": "chat",
|
||||
"modalities": ["text"],
|
||||
"pricing": {"prompt_token_price": 0.0, "completion_token_price": 0.0, "currency": "CR",
|
||||
"free_until": "2026-12-01"},
|
||||
"context_length": 524288, "default_max_output_tokens": 16384,
|
||||
"capabilities": {
|
||||
"tools": True, "thinking": True, "reasoning_effort": True, "reasoning_effort_values": None,
|
||||
"reasoning_effort_aliases": {"medium": "high", "minimal": "low"}, "vision": False,
|
||||
"max_images_per_request": None, "response_format": True, "seed": True, "top_k": True, "stop": True,
|
||||
},
|
||||
"precision": "FP8",
|
||||
}
|
||||
|
||||
|
||||
def test_models_capability_fields():
|
||||
body = {"data": [GLM_53_DETAIL]}
|
||||
|
||||
m = _client(lambda req: httpx.Response(200, json=body)).models()[0]
|
||||
|
||||
assert m.context_length == 524288 and m.default_max_output_tokens == 16384 and m.precision == "FP8"
|
||||
assert m.capabilities.tools is True and m.capabilities.thinking is True
|
||||
assert m.capabilities.reasoning_effort is True and m.capabilities.reasoning_effort_values is None
|
||||
assert m.capabilities.reasoning_effort_aliases == {"medium": "high", "minimal": "low"}
|
||||
assert m.capabilities.vision is False and m.capabilities.max_images_per_request is None
|
||||
assert m.capabilities.response_format is True
|
||||
assert m.capabilities.seed is True and m.capabilities.top_k is True and m.capabilities.stop is True
|
||||
|
||||
|
||||
def test_get_model():
|
||||
seen = {}
|
||||
|
||||
def h(req):
|
||||
seen["path"] = req.url.path
|
||||
return httpx.Response(200, json=GLM_53_DETAIL)
|
||||
|
||||
m = _client(h).get_model("glm-5.3")
|
||||
|
||||
assert seen["path"] == "/v1/models/glm-5.3"
|
||||
assert m.id == "glm-5.3" and m.context_length == 524288 and m.precision == "FP8"
|
||||
assert m.capabilities.tools is True and m.capabilities.vision is False
|
||||
|
||||
|
||||
def test_get_model_not_found():
|
||||
resp = httpx.Response(404, json={"error": {"message": "model bulunamadı: yok-xyz",
|
||||
"type": "invalid_request_error", "code": "model_not_found"}})
|
||||
with pytest.raises(LLMNotFoundError, match="bulunamadı"):
|
||||
_client(lambda req: resp).get_model("yok-xyz")
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_async_get_model():
|
||||
c = AsyncEvrenLLMClient(KEY, transport=httpx.MockTransport(lambda req: httpx.Response(200, json=GLM_53_DETAIL)))
|
||||
async with c:
|
||||
m = await c.get_model("glm-5.3")
|
||||
assert m.id == "glm-5.3" and m.capabilities.reasoning_effort is True and m.default_max_output_tokens == 16384
|
||||
|
||||
|
||||
def test_validation_error():
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user