commoncompute 0.1.0__tar.gz → 0.1.2__tar.gz

{commoncompute-0.1.0 → commoncompute-0.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: commoncompute
-Version: 0.1.0
+Version: 0.1.2
 Summary: Official Python SDK for Common Compute — the batch AI bill you shouldn't be paying.
 Project-URL: Homepage, https://commoncompute.ai
 Project-URL: Documentation, https://commoncompute.ai/docs
@@ -61,7 +61,8 @@ import commoncompute as cc
 
 client = cc.Client()   # CC_API_KEY, or ~/.commoncompute/config
 
-job = client.transcription.create("s3://my-bucket/meeting.mp3", timestamps=True)
+job = client.submit("coreml_embed", {"input": ["what is the neural engine?"]},
+                    model_id="bge-base")
 print(job.job_id, job.locked_price_usd, job.eta_seconds)   # price known BEFORE it runs
 
 result = client.result(job)          # waits, downloads, attaches the receipt
@@ -69,7 +70,10 @@ print(result.output)
 print(result.receipt)                # signed proof of what ran, where, for how much
 ```
 
-## Examples
+Prefer the OpenAI-style surface? `client.embeddings.create(input=[...],
+model="bge-base")` returns vectors synchronously.
+
+## Examples — live today
 
 ### 1. OCR / document extraction — vs AWS Textract
 
@@ -78,38 +82,40 @@ job = client.ocr.extract("https://example.com/contracts.pdf", structured=True)
 result = client.result(job)          # blocks + bounding boxes as JSON
 ```
 
-### 2. Transcription in bulk
+### 2. Embeddings in bulk
 
 ```python
-jobs = client.transcription.create_many(
-    [f"s3://recordings/call-{i}.mp3" for i in range(200)],
-    language="en",
+for result in client.submit_many(
+    "coreml_embed",
+    [{"input": chunked(doc)} for doc in corpus],
     max_concurrent=16,
-)
+):
+    save(result.output)              # results stream as they complete
 ```
 
-### 3. Reranking — sharpen RAG retrieval
+### 3. Background removal
 
 ```python
-job = client.rerank("what is the neural engine?", documents, top_n=5)
-top = client.result(job).output
+job = client.images.remove_background("product-shot.png")
 ```
 
-### 4. Translation
+Also live: speech synthesis, image classification/detection/pose, barcode
+reading, HEIC conversion — plus the generic `client.submit(workload_id,
+payload)` for anything marked **live** in
+[the catalog](https://commoncompute.ai/workloads).
 
-```python
-job = client.translate(catalog_descriptions, target_lang="de")
-```
+### In preview
 
-### 5. Background removal
+`client.translate(...)` and `client.video.transcode(...)` run on preview
+capacity — functional, not yet GA.
 
-```python
-job = client.images.remove_background("product-shot.png")
-```
+### Not live yet
 
-Also available: `client.video.transcode(...)`, `client.build.ios_test(...)`,
-`client.images.generate(...)`, `client.embeddings.create(...)`, and the
-generic `client.submit(workload_id, payload)` for anything in
+`client.transcription.*`, `client.rerank(...)`, `client.images.generate(...)`,
+`client.chat.*`, and `client.build.ios_test(...)` target workloads still
+marked *coming soon*: the API refuses them with a clear
+`workload_not_available` error (HTTP 409) rather than queueing a job that
+won't run. They activate automatically as those lanes go live — watch
 [the catalog](https://commoncompute.ai/workloads).
 
 ## Price before execution, hard caps, dry runs

{commoncompute-0.1.0 → commoncompute-0.1.2}/README.md

@@ -26,7 +26,8 @@ import commoncompute as cc
 
 client = cc.Client()   # CC_API_KEY, or ~/.commoncompute/config
 
-job = client.transcription.create("s3://my-bucket/meeting.mp3", timestamps=True)
+job = client.submit("coreml_embed", {"input": ["what is the neural engine?"]},
+                    model_id="bge-base")
 print(job.job_id, job.locked_price_usd, job.eta_seconds)   # price known BEFORE it runs
 
 result = client.result(job)          # waits, downloads, attaches the receipt
@@ -34,7 +35,10 @@ print(result.output)
 print(result.receipt)                # signed proof of what ran, where, for how much
 ```
 
-## Examples
+Prefer the OpenAI-style surface? `client.embeddings.create(input=[...],
+model="bge-base")` returns vectors synchronously.
+
+## Examples — live today
 
 ### 1. OCR / document extraction — vs AWS Textract
 
@@ -43,38 +47,40 @@ job = client.ocr.extract("https://example.com/contracts.pdf", structured=True)
 result = client.result(job)          # blocks + bounding boxes as JSON
 ```
 
-### 2. Transcription in bulk
+### 2. Embeddings in bulk
 
 ```python
-jobs = client.transcription.create_many(
-    [f"s3://recordings/call-{i}.mp3" for i in range(200)],
-    language="en",
+for result in client.submit_many(
+    "coreml_embed",
+    [{"input": chunked(doc)} for doc in corpus],
     max_concurrent=16,
-)
+):
+    save(result.output)              # results stream as they complete
 ```
 
-### 3. Reranking — sharpen RAG retrieval
+### 3. Background removal
 
 ```python
-job = client.rerank("what is the neural engine?", documents, top_n=5)
-top = client.result(job).output
+job = client.images.remove_background("product-shot.png")
 ```
 
-### 4. Translation
+Also live: speech synthesis, image classification/detection/pose, barcode
+reading, HEIC conversion — plus the generic `client.submit(workload_id,
+payload)` for anything marked **live** in
+[the catalog](https://commoncompute.ai/workloads).
 
-```python
-job = client.translate(catalog_descriptions, target_lang="de")
-```
+### In preview
 
-### 5. Background removal
+`client.translate(...)` and `client.video.transcode(...)` run on preview
+capacity — functional, not yet GA.
 
-```python
-job = client.images.remove_background("product-shot.png")
-```
+### Not live yet
 
-Also available: `client.video.transcode(...)`, `client.build.ios_test(...)`,
-`client.images.generate(...)`, `client.embeddings.create(...)`, and the
-generic `client.submit(workload_id, payload)` for anything in
+`client.transcription.*`, `client.rerank(...)`, `client.images.generate(...)`,
+`client.chat.*`, and `client.build.ios_test(...)` target workloads still
+marked *coming soon*: the API refuses them with a clear
+`workload_not_available` error (HTTP 409) rather than queueing a job that
+won't run. They activate automatically as those lanes go live — watch
 [the catalog](https://commoncompute.ai/workloads).
 
 ## Price before execution, hard caps, dry runs

{commoncompute-0.1.0 → commoncompute-0.1.2}/commoncompute/__init__.py

@@ -11,33 +11,26 @@ Quickstart::
     client = cc.Client(api_key="cc_live_...")
     print(client.balance())
 
-    # OpenAI-compatible chat
-    reply = client.chat.create(
-        model="qwen-2.5-7b-instruct",
-        messages=[{"role": "user", "content": "hi"}],
-    )
-    print(reply["choices"][0]["message"]["content"])
-
-    # Native job submission
+    # Native job submission (embeddings — live today)
     job = client.jobs.submit(
         workload_id="coreml_embed",
-        payload={"inputs": ["hello world"]},
+        payload={"input": ["hello world"]},
     )
     result = client.jobs.wait(job["id"])
 
-Streaming chat::
+    # OpenAI-compatible embeddings
+    vectors = client.embeddings.create(input=["hello world"], model="bge-base")
 
-    for chunk in client.chat.stream(model="qwen-2.5-7b-instruct",
-                                    messages=[{"role": "user", "content": "hi"}]):
-        delta = chunk["choices"][0].get("delta", {}).get("content", "")
-        print(delta, end="", flush=True)
+`client.chat.*` targets LLM workloads that are still marked *coming soon*
+in the catalog — the API declines them with `workload_not_available`
+(HTTP 409) until they go live. See https://commoncompute.ai/workloads.
 """
 
 # MUST come before any submodule imports — _transport.py reads
 # `commoncompute.__version__` at import time for the User-Agent header,
 # and the previous ordering (`__version__` defined at the bottom) made
 # every clean install of the SDK fail with a circular ImportError.
-__version__ = "0.1.0"
+__version__ = "0.1.2"
 
 from .client import Client, AsyncClient
 from .errors import (

{commoncompute-0.1.0 → commoncompute-0.1.2}/commoncompute/_generated_workloads.py

@@ -25,6 +25,19 @@ KNOWN_WORKLOAD_IDS: frozenset[str] = frozenset({
     "vision_face",
     "vision_detect",
     "vision_pose",
+    "vision_docseg",
+    "vision_rectangles",
+    "vision_hand_pose",
+    "vision_animal_pose",
+    "vision_saliency",
+    "vision_animals",
+    "vision_classify",
+    "vision_person_seg",
+    "vision_trajectory",
+    "vision_optical_flow",
+    "av_compose",
+    "pdfkit_ops",
+    "createml_train",
     "vision_barcode",
     "vision_segment",
     "coreml_upscale",
@@ -43,6 +56,16 @@ KNOWN_WORKLOAD_IDS: frozenset[str] = frozenset({
     "mlx_finetune",
     "pytorch_mps",
     "blender_metal",
+    "speech_transcribe",
+    "mlx_vlm",
+    "mlx_llm_cluster",
+    "foundation_llm",
+    "vision_similarity",
+    "image_filter",
+    "audio_normalize",
+    "webkit_test",
+    "notarize",
+    "coreml_convert",
     "cpu_bench",
 })
 

{commoncompute-0.1.0 → commoncompute-0.1.2}/commoncompute/_transport.py

@@ -27,7 +27,10 @@ from . import __version__
 from .errors import CommonComputeError, NetworkError, _from_status
 
 DEFAULT_BASE_URL = "https://api.commoncompute.ai"
-DEFAULT_TIMEOUT = 60.0
+# Above the API's 120s synchronous poll windows (chat/images block server-side
+# up to 120s) — a 60s default made every slow-but-successful sync call time
+# out client-side and retry into an idempotent-replay 202.
+DEFAULT_TIMEOUT = 180.0
 DEFAULT_MAX_RETRIES = 2
 
 
@@ -40,8 +43,14 @@ def _key_from_config_files() -> Optional[str]:
     """
     import pathlib
 
+    base = pathlib.Path(os.environ.get("CC_CONFIG_DIR") or (pathlib.Path.home() / ".commoncompute"))
     candidates = [
-        pathlib.Path(os.environ.get("CC_CONFIG_DIR") or (pathlib.Path.home() / ".commoncompute")) / "config",
+        base / "config",
+        # connect()/`cc login` save here (respects CC_CONFIG_DIR) — the
+        # transport previously only checked the hard-coded legacy path, so a
+        # key saved under a custom CC_CONFIG_DIR was invisible to new
+        # processes.
+        base / "credentials",
         pathlib.Path.home() / ".config" / "commoncompute" / "credentials",
     ]
     for path in candidates:
@@ -87,10 +96,13 @@ def _retry_delay(attempt: int, response: Optional[httpx.Response]) -> float:
         ra = response.headers.get("retry-after")
         if ra:
             try:
-                return min(max(float(ra), 0.0), 30.0)
+                # Honor the server's window up to 60s — the API sends
+                # Retry-After: 60 on 429s, and clamping to 30 guaranteed the
+                # retry landed inside the same rate-limit window.
+                return min(max(float(ra), 0.0), 60.0)
             except ValueError:
                 pass
-    # 0.5s, 1.5s (+ jitter) — bounded so 3 attempts fit a 60s caller budget.
+    # 0.5s, 1.5s (+ jitter).
     return 0.5 * (3**attempt) + random.random() * 0.25
 
 
@@ -203,6 +215,7 @@ class SyncTransport(_BaseTransport):
         data: Any = None,
         files: Any = None,
         headers: Optional[Mapping[str, str]] = None,
+        raw: bool = False,
     ) -> Any:
         hdrs = _build_headers(self._api_key, self._org, headers)
         # Fresh Idempotency-Key per logical call makes retried POSTs replay
@@ -233,7 +246,9 @@ class SyncTransport(_BaseTransport):
                 time.sleep(_retry_delay(attempt, resp))
                 continue
             _raise_for(resp)
-            return _parse_body(resp)
+            # raw=True hands back the httpx.Response so binary bodies
+            # (media results) aren't UTF-8-mangled by the text fallback.
+            return resp if raw else _parse_body(resp)
         if isinstance(last_exc, httpx.TransportError):
             raise NetworkError(str(last_exc), code="network_error") from last_exc
         raise last_exc if last_exc else CommonComputeError("request failed", status_code=None, code="request_failed")

{commoncompute-0.1.0 → commoncompute-0.1.2}/commoncompute/cli.py

@@ -431,7 +431,7 @@ def keys_revoke(key_id: str) -> None:
 @app.command()
 def chat(
     prompt: str = typer.Argument(..., help="Your message."),
-    model: str = typer.Option("qwen-2.5-7b-instruct", "--model", "-m"),
+    model: str = typer.Option("qwen-2.5-7b", "--model", "-m"),
     stream: bool = typer.Option(True, "--stream/--no-stream"),
 ) -> None:
     """Send a one-shot chat message to a model."""
@@ -439,7 +439,14 @@ def chat(
     try:
         if stream:
             for chunk in c.chat.stream(model=model, messages=[{"role": "user", "content": prompt}]):
-                delta = chunk.get("choices", [{}])[0].get("delta", {}).get("content", "") or ""
+                choice = chunk.get("choices", [{}])[0]
+                # delta-shaped per the OpenAI stream contract; fall back to a
+                # message-shaped chunk (older deployments sent those).
+                delta = (
+                    choice.get("delta", {}).get("content", "")
+                    or choice.get("message", {}).get("content", "")
+                    or ""
+                )
                 sys.stdout.write(delta)
                 sys.stdout.flush()
             sys.stdout.write("\n")

{commoncompute-0.1.0 → commoncompute-0.1.2}/commoncompute/client.py

@@ -14,7 +14,7 @@ from pathlib import Path
 from typing import Any, AsyncIterator, BinaryIO, Iterator, Mapping, Optional, Sequence, Union
 
 from ._transport import AsyncTransport, SyncTransport
-from .errors import JobTimeoutError, NotFoundError
+from .errors import CommonComputeError, JobTimeoutError, NotFoundError
 from .models import Job, Quote, Receipt, TaskResult
 from .tasks import (
     OCR,
@@ -114,19 +114,36 @@ class _JobsSync:
 
     def download(self, task_id: str, *, dest: Optional[Union[str, Path]] = None) -> Union[bytes, Path]:
         """Download the job result. Returns bytes, or the path if ``dest`` is given."""
-        data = self._t.request("GET", f"/v1/jobs/download/{task_id}")
-        if isinstance(data, dict):
-            # The API may return a pre-signed URL instead of inlined bytes.
-            url = data.get("url") or data.get("download_url")
-            if url:
+        # Raw-bytes fetch of GET /v1/jobs/:id/result: the parsed-transport
+        # path UTF-8-mangled binary media outputs, and a 202 "Result not
+        # ready" body was silently saved to disk as the result.
+        resp = self._t.request("GET", f"/v1/jobs/{task_id}/result", raw=True)
+        if resp.status_code == 202:
+            state = "unknown"
+            try:
+                state = resp.json().get("state", "unknown")
+            except Exception:
+                pass
+            raise CommonComputeError(
+                f"Result not ready for job {task_id} (state={state}) — wait() for completion first.",
+                status_code=202,
+                code="result_not_ready",
+            )
+        content_type = resp.headers.get("content-type", "")
+        blob: bytes
+        if "application/json" in content_type:
+            try:
+                data = resp.json()
+            except ValueError:
+                data = None
+            if isinstance(data, dict) and (data.get("url") or data.get("download_url")):
+                # The API may return a pre-signed URL instead of inlined bytes.
                 import httpx as _httpx
-                blob = _httpx.get(url, timeout=60.0).content
+                blob = _httpx.get(data.get("url") or data.get("download_url"), timeout=60.0).content
             else:
-                blob = str(data).encode()
-        elif isinstance(data, (bytes, bytearray)):
-            blob = bytes(data)
+                blob = resp.content
         else:
-            blob = str(data).encode()
+            blob = resp.content
         if dest is None:
             return blob
         dest_path = Path(dest)

{commoncompute-0.1.0 → commoncompute-0.1.2}/commoncompute/tasks.py

@@ -230,7 +230,7 @@ class Images(_TaskBase):
         self,
         prompt: str,
         *,
-        model: str = "flux-schnell",
+        model: Optional[str] = None,
         n: int = 1,
         size: Optional[str] = None,
         max_spend_usd: Optional[float] = None,
@@ -239,7 +239,12 @@ class Images(_TaskBase):
         payload: dict = {"prompt": prompt, "n": n}
         if size:
             payload["size"] = size
-        return self._run("mlx_image", payload, units=n, max_spend_usd=max_spend_usd, dry_run=dry_run)
+        # model was previously accepted and silently dropped (and its old
+        # default "flux-schnell" belongs to sd_mlx, not mlx_image) — forward
+        # it so the server can route/validate it.
+        return self._run(
+            "mlx_image", payload, units=n, model_id=model, max_spend_usd=max_spend_usd, dry_run=dry_run
+        )
 
 
 class Video(_TaskBase):

{commoncompute-0.1.0 → commoncompute-0.1.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "commoncompute"
-version = "0.1.0"
+version = "0.1.2"
 description = "Official Python SDK for Common Compute — the batch AI bill you shouldn't be paying."
 readme = "README.md"
 requires-python = ">=3.9"