repo_name stringlengths 1 62 | dataset stringclasses 1
value | lang stringclasses 11
values | pr_id int64 1 20.1k | owner stringlengths 2 34 | reviewer stringlengths 2 39 | diff_hunk stringlengths 15 262k | code_review_comment stringlengths 1 99.6k |
|---|---|---|---|---|---|---|---|
openrouter-runner | github_2023 | python | 68 | OpenRouterTeam | alexanderatallah | @@ -70,70 +82,56 @@ def __init__(self, params: VllmParams):
async def generate(self, payload: CompletionPayload, params):
assert self.engine is not None, "Engine not initialized"
+ # Track usage as a running total
+ # NOTE: This does NOT yet include cold-start GPU time | how hard is this to add @sambarnes ? |
openrouter-runner | github_2023 | python | 68 | OpenRouterTeam | louisgv | @@ -95,45 +100,29 @@ class Usage(BaseModel):
prompt_tokens: int
completion_tokens: int
- # TODO: add these in after deprecating the old fields
- # duration: float
- # gpu_type: GPUType
- # gpu_count: int
+ duration: float
+ gpu_type: GPUType
+ gpu_count: int
class ResponseBody(Base... | The main reason we used this was due to the pydantic version shipped with vllm - it's deprecated in latest (due to modal installation), but not if we have the vllm's pydantic typing xd.....
I wanted to do this way back in September when I saw the red deprecation strike! - Let's try it out on dev and see if it work! |
openrouter-runner | github_2023 | python | 68 | OpenRouterTeam | louisgv | @@ -70,84 +84,82 @@ def __init__(self, params: VllmParams):
async def generate(self, payload: CompletionPayload, params):
assert self.engine is not None, "Engine not initialized"
- t_start_inference = time.perf_counter()
+ # Track usage as a running total. For the first request to the
+ ... | @sambarnes we need to yield "something" to keep the connection going, based on an experiment I did some time ago with fetch + SSE. Basically a hack to keep the connection going since this is a one-way pull, not a bi-directional connection like websocket where there's a dedicated handshake before transmission.
It's k... |
openrouter-runner | github_2023 | python | 68 | OpenRouterTeam | louisgv | @@ -70,84 +84,82 @@ def __init__(self, params: VllmParams):
async def generate(self, payload: CompletionPayload, params):
assert self.engine is not None, "Engine not initialized"
- t_start_inference = time.perf_counter()
+ # Track usage as a running total. For the first request to the
+ ... | Feel like we don't have to reset it here. The exception could have been thrown mid generation :-? |
openrouter-runner | github_2023 | python | 39 | OpenRouterTeam | louisgv | @@ -29,3 +29,12 @@ async def test_completion_streaming():
output = await completion(prompt, stream=True)
assert output is not None
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_large_prompt():
+ """Should accept & complete a large prompt"""
+ prompt = get_words_from_file(n=2500)
... | Ooh interesting - how else and what other aspect of an LLM endpoint would you test from an e2e? |
openrouter-runner | github_2023 | python | 64 | OpenRouterTeam | alexanderatallah | @@ -85,10 +85,28 @@ class CompletionPayload(BaseModel):
runner: RunnerConfiguration | None = None
-class ResponseBody(BaseModel):
- text: str
+class Usage(BaseModel):
+ """
+ An OpenAI-style usage struct, containing the expected
+ token counts as well as additional data about the GPU
+ usage of ... | yeah it seems redundant to me. this repo doesn't follow their spec anyway, so I think we could omit it |
openrouter-runner | github_2023 | python | 64 | OpenRouterTeam | alexanderatallah | @@ -87,22 +88,29 @@ async def generate(self, payload: CompletionPayload, params):
and request_output.outputs[0].text[-1] == "\ufffd"
):
continue
- token = request_output.outputs[0].text[index:]
- index = len(req... | OpenAI doesn't do any accounting for streaming :(
We should, by sending the running token counts along with the running total time |
openrouter-runner | github_2023 | python | 61 | OpenRouterTeam | sambarnes | @@ -15,11 +15,20 @@
def completion(
+ request: Request,
payload: CompletionPayload,
):
model_path = get_model_path(payload.model)
logger.info(
- "Received completion request", extra={"model": str(model_path)}
+ "Received completion request",
+ extra={ | IIRC at opensea we did something that set this context globally for a request so all logs in the call stack used it 🤔 but that's just something in the back of my mind, we can do it later if necessary |
openrouter-runner | github_2023 | python | 61 | OpenRouterTeam | sambarnes | @@ -15,11 +15,20 @@
def completion(
+ request: Request,
payload: CompletionPayload,
):
model_path = get_model_path(payload.model)
logger.info(
- "Received completion request", extra={"model": str(model_path)}
+ "Received completion request",
+ extra={
+ "model": s... | I forget if this key needs to be snake_case for datadog to pick it up |
openrouter-runner | github_2023 | python | 60 | OpenRouterTeam | alexanderatallah | @@ -1,9 +1,12 @@
from enum import Enum
-from typing import List, Optional, Union
+from typing import Final, List, Optional, Union
from fastapi.responses import JSONResponse, PlainTextResponse
from pydantic import BaseModel
+A100_40G: Final[float] = 0.001036
+A100_80G: Final[float] = 0.001553 |
can we name this variable based on the units? ($/sec?) |
openrouter-runner | github_2023 | python | 53 | OpenRouterTeam | alexanderatallah | @@ -77,11 +77,9 @@ async def generate(self, payload: CompletionPayload, params):
try:
import time
- tags = {"model": self.engine_args.model}
- with timer("engine.generate()", tags=tags): | we're removing this bc it was instant right? |
openrouter-runner | github_2023 | python | 49 | OpenRouterTeam | montasaurus | @@ -31,6 +33,17 @@ def add_observability(image: Image):
)
+@contextmanager
+def timer(action: str, tags: dict[str, str | int] = None) -> None:
+ """A simple timer context manager with structured logging for its output."""
+ start = time.perf_counter()
+ yield
+ elapsed = time.perf_counter() - start... | ```suggestion
extra = (tags or {}) | {"duration": elapsed}
```
to fix the precedence |
openrouter-runner | github_2023 | others | 51 | OpenRouterTeam | louisgv | @@ -0,0 +1,39 @@
+name: Release (prod)
+
+on:
+ push:
+ branches: [main]
+ paths: ['modal/**'] | We should add the poetry lock file as well |
openrouter-runner | github_2023 | others | 51 | OpenRouterTeam | louisgv | @@ -0,0 +1,39 @@
+name: Release (prod)
+
+on:
+ push:
+ branches: [main]
+ paths: ['modal/**']
+
+jobs:
+ release:
+ name: Release (prod)
+ runs-on: ubuntu-latest
+ env:
+ MODAL_TOKEN_ID: ${{ secrets.MODAL_TOKEN_ID }}
+ MODAL_TOKEN_SECRET: ${{ secrets.MODAL_TOKEN_SECRET }} | Done! cc @alexanderatallah to config @sambarnes 's perm |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | sambarnes | @@ -53,7 +58,7 @@ def download_model(model_name: str):
ignore_patterns.append("*.bin")
# Clean doesn't remove the cache, so using `local_files_only` here returns the cache even when the local dir is empty.
- print(f"Checking for {model_name}")
+ logger.info(f"Checking for {model_name}") | in the future we might wanna move more towards structured logging with some of these
that way we can e.g. group logs by model in DD
`logger.info("Checking for model", extra=dict(model=model_name)"`
def not a right now thing, but worth considering going forward
|
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | sambarnes | @@ -0,0 +1,107 @@
+import json
+import logging
+import os
+import sys
+
+from datadog_api_client import ApiClient, Configuration
+from datadog_api_client.v2.api.logs_api import LogsApi
+from datadog_api_client.v2.model.content_encoding import ContentEncoding
+from datadog_api_client.v2.model.http_log import HTTPLog
+fr... | given the new docs say:
> or leave it blank to disable Datadog (e.g. `DD_API_KEY=`
do you know if `DD_API_KEY` will still be on the container as empty string if ppl set it like that?
if so, would probs need to be
```suggestion
if os.environ.get("DD_API_KEY") is not None:
``` |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | sambarnes | @@ -0,0 +1,107 @@
+import json
+import logging
+import os
+import sys
+
+from datadog_api_client import ApiClient, Configuration
+from datadog_api_client.v2.api.logs_api import LogsApi
+from datadog_api_client.v2.model.content_encoding import ContentEncoding
+from datadog_api_client.v2.model.http_log import HTTPLog
+fr... | think we should still do a bare `logging.exception(` in order for the sentry hook to fire & notify us there?
```suggestion
logging.exception("Error sending log to Datadog")
``` |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | sambarnes | @@ -0,0 +1,8 @@
+from modal import Image
+
+
+def add_datadog(image: Image):
+ return image.pip_install("datadog-api-client==2.21.0") | mind expanding one why this helper would have api & not ddtrace in it?
(just mentioning cause i see ddtrace manually on the download.py image) |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | sambarnes | @@ -0,0 +1,65 @@
+from fastapi import Depends, FastAPI, Request
+from fastapi.responses import JSONResponse
+from fastapi.security import HTTPAuthorizationCredentials
+from pydantic import BaseModel
+
+from runner.endpoints.completion import completion as completion_endpoint
+from runner.shared.common import config
+fr... | i like this reorganization :+1: |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | louisgv | @@ -11,13 +13,16 @@
models_volume,
)
+logger = get_logger(__name__) | Does this work at the top-level scope? Does the stub's env injection take effect here? |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | louisgv | @@ -11,13 +13,16 @@
models_volume,
)
+logger = get_logger(__name__)
+
cache_path = get_model_path("__cache__")
downloader_image = (
- Image.debian_slim()
+ BASE_IMAGE
# Use the barebones hf-transfer package for maximum download speeds. No progress bar, but expect 700MB/s.
.pip_install("huggi... | This would install on top of the `BASE_IMAGE` instance which is not what we want I think. We should make BASE_IMAGE into a function that return a new Image() instance instead. |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | louisgv | @@ -0,0 +1,107 @@
+import json
+import logging
+import os
+import sys
+
+from datadog_api_client import ApiClient, Configuration
+from datadog_api_client.v2.api.logs_api import LogsApi
+from datadog_api_client.v2.model.content_encoding import ContentEncoding
+from datadog_api_client.v2.model.http_log import HTTPLog
+fr... | Is this the case even for prod instance? |
openrouter-runner | github_2023 | python | 48 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,80 @@
+import modal
+from fastapi import Depends, FastAPI, Request
+from fastapi.responses import JSONResponse
+from fastapi.security import HTTPAuthorizationCredentials
+from pydantic import BaseModel
+
+from runner.endpoints.completion import completion as completion_endpoint
+from runner.shared.common imp... | when is this called? |
openrouter-runner | github_2023 | python | 47 | OpenRouterTeam | montasaurus | @@ -116,7 +117,9 @@ async def generate(self, payload: CompletionPayload, params):
print(f"Request completed: {throughput:.4f} tokens/s")
except Exception as err:
e = create_error_text(err)
- print(e)
+ logging.exception(
+ "Failed generation", extr... | I hit the same thing, `logging.basicConfig(level=logging.INFO)` fixed it |
openrouter-runner | github_2023 | python | 46 | OpenRouterTeam | montasaurus | @@ -70,7 +72,13 @@ async def get_job(
@stub.function(
- secret=Secret.from_name("ext-api-key"),
+ image=Image.debian_slim(python_version="3.10").pip_install(
+ "sentry-sdk[fastapi]==1.39.1"
+ ), | Got exactly that in a PR I'm working on! |
openrouter-runner | github_2023 | python | 43 | OpenRouterTeam | louisgv | @@ -10,7 +10,9 @@ def get_model_path(model_name: str):
return models_path / model_name.lower()
-def does_model_exist(model_path: Path):
+# gets passed a str in some circumstances despite being typed as Path
+def does_model_exist(model_path: Path | str):
+ model_path = Path(model_path) | Might need to revert this (and the model_path passing down to our container) to just string. When we instantiate a GPU container, the props passed down will be serialized by the Modal abstraction, using pickle, and afaik it will not be deserialized on our code properly.
3 options:
1. Ping Modal about better type... |
openrouter-runner | github_2023 | typescript | 41 | OpenRouterTeam | louisgv | @@ -0,0 +1,35 @@
+import {
+ awaitJob,
+ enqueueAddModel,
+ getApiUrl,
+ getAuthHeaders,
+ runIfCalledAsScript
+} from 'scripts/shared';
+
+async function main(model?: string) {
+ const modelName = model || process.env.MODEL!;
+ console.log(`Test adding model ${modelName}`);
+ const body = await enqueueAddModel... | nit: Can we think of an alternative name that doesn't repeat the `await`? |
openrouter-runner | github_2023 | python | 41 | OpenRouterTeam | louisgv | @@ -24,7 +25,7 @@ def _make_container(
class _VllmContainer(VllmEngine):
def __init__(
self,
- model_path: str,
+ model_path: Path, | Interesting - so it's possible to use `Path` as a param such that only 1 instance will be spawned to serve a certain path until the max concurrent input is reached? |
openrouter-runner | github_2023 | typescript | 41 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,34 @@
+import {
+ completion,
+ enqueueAddModel,
+ pollForJobCompletion,
+ runIfCalledAsScript
+} from 'scripts/shared';
+
+async function main() {
+ const modelName = 'microsoft/Orca-2-13b'; | can we do phi-2? (if this test is still helpful for small models)
it has a better license |
openrouter-runner | github_2023 | python | 41 | OpenRouterTeam | alexanderatallah | @@ -24,12 +37,17 @@ class Params(BaseModel):
skip_special_tokens: bool = True
-class Payload(BaseModel):
+class RunnerConfiguration(BaseModel):
+ container: ContainerType
+
+
+class CompletionPayload(BaseModel):
id: str
prompt: str
stream: bool = False
params: Params
model: str
+ ... | woah, new union notation |
openrouter-runner | github_2023 | typescript | 41 | OpenRouterTeam | louisgv | @@ -6,9 +6,21 @@ const envFile = `.env.dev`;
config({ path: envFile });
-const url = process.env.API_URL;
-const key = process.env.RUNNER_API_KEY;
+const url = process.env.API_URL!;
+const key = process.env.RUNNER_API_KEY!; | We should assert these instead of using `!` since new devs might not have them. We lean toward strict typing with assertion as much as possible. |
openrouter-runner | github_2023 | typescript | 41 | OpenRouterTeam | louisgv | @@ -0,0 +1,35 @@
+import {
+ enqueueAddModel,
+ getApiUrl,
+ getAuthHeaders,
+ pollForJobCompletion,
+ runIfCalledAsScript
+} from 'scripts/shared';
+
+async function main(model?: string) {
+ const modelName = model || process.env.MODEL!;
+ console.log(`Test adding model ${modelName}`);
+ const body = await enq... | nit: we should prob reuse the variable from shared |
openrouter-runner | github_2023 | python | 33 | OpenRouterTeam | louisgv | @@ -22,6 +22,13 @@ def download_models(all_models: List[str]):
cache_path = get_model_path("__cache__")
for model_name in all_models:
+ parts = model_name.split(":")
+
+ model_revision = None | This only downloads the model - can we also added it somehow into the protocol's input params? |
openrouter-runner | github_2023 | python | 36 | OpenRouterTeam | alexanderatallah | @@ -28,27 +27,39 @@ def __init__(
model_path: str,
max_model_len: Optional[int] = None,
):
- if num_gpus > 1:
- # Patch issue from https://github.com/vllm-project/vllm/issues/1116
- import ray
-
- ray.shutdown()
- ... | we can't init sentry higher up, for other Modal processes too? |
openrouter-runner | github_2023 | others | 32 | OpenRouterTeam | louisgv | @@ -48,3 +48,11 @@ ignore = [
[tool.ruff.per-file-ignores]
"gcp/shap-e/app.py" = ["E402"] # module level import not at top of file -- expected for sys.path.insert() usage
+[tool.ruff.lint.isort]
+known-third-party = ["modal"]
+known-first-party = [
+ "punctuator",
+ "runner",
+ "shared",
+ "tuner",
+] | I wonder if there's a better way to structure our project? Can ruff target or isort has sub projects? |
openrouter-runner | github_2023 | python | 31 | OpenRouterTeam | louisgv | @@ -0,0 +1,29 @@
+import os
+from typing import Annotated
+
+from fastapi import Depends, FastAPI, testclient
+from fastapi.security import HTTPAuthorizationCredentials
+
+from shared.config import Config
+
+
+def test_auth():
+ """The API auth dependency should prevent unauthorized requests.""" | @sambarnes I think the CI PR targeted the wrong remote (or wrong github org) |
openrouter-runner | github_2023 | python | 31 | OpenRouterTeam | louisgv | @@ -75,16 +72,12 @@ def transform(self, input_str: str):
import threading
import time
- output = [
- None
- ] # Use a list to hold the output to bypass Python's scoping limitations | Ooh - I also forgot about the line-length. Kinda want to keep it at 80 since I split my screen a lot :d |
openrouter-runner | github_2023 | others | 31 | OpenRouterTeam | louisgv | @@ -16,11 +16,13 @@ accelerate = "^0.23.0"
datasets = "^2.14.5"
scipy = "^1.11.3"
wandb = "^0.15.12"
+httpx = "^0.26.0" | seems like we don't need this anymore? Based on your comment, I think this was needed for the e2e test.
For e2e, I wonder if we can do a `modal deply runner` to a `e2e` workspace, tail the modal URL, then call it with our JS scripts?
Also we should prob add a js script to test auth as well :-? |
openrouter-runner | github_2023 | typescript | 31 | OpenRouterTeam | louisgv | @@ -15,6 +15,17 @@ async function main(model?: string) {
max_tokens: 1024,
stop: ['</s>'],
});
+
+ // Unauthorized requests should fail with a 401
+ let gotExpectedError = false;
+ try {
+ await completion(prompt, {model, apiKey: "BADKEY"});
+ } catch (e: any) {
+ gotExpectedError = e.message =... | nit: we can just re-throw the error in the try if not 401, no need to do it separately IMO. Also, we should either use `startsWith` or strict equal. |
openrouter-runner | github_2023 | others | 28 | OpenRouterTeam | louisgv | @@ -1,46 +1,260 @@
-# PREREQ:
+# OpenRouter Runner - Setup Guide
-1. Create a modal secret group
- `HUGGINGFACE_TOKEN = <your huggingface token>`
- with name "huggingface"
-2. Create a modal secret group
- `RUNNER_API_KEY = <generate a random key>`
- with name "ext-api-key"
-3. Make sure your current d... | nit: the official casing is `vLLM` |
openrouter-runner | github_2023 | python | 29 | OpenRouterTeam | louisgv | @@ -89,7 +90,7 @@ async def generate(self, payload: Payload, params):
async for request_output in results_generator:
final_output = request_output
- output = request_output.outputs[0].text | Good catch - will run the e2e test without stream on this path, but this diff makes sense to me |
openrouter-runner | github_2023 | python | 17 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,54 @@
+from runner.engines.vllm import VllmEngine, VllmParams
+
+# Each container comprises of:
+# 1. An image
+# 2. A stub class wrapping an engine
+
+from modal import gpu, Image
+
+from shared.volumes import models_path
+from runner.shared.common import stub
+
+_gpu = gpu.A100(count=1, memory=80)
+
+_vllm... | do we need this? seems like vllm merge da pr sept 27 to remove the need: https://github.com/vllm-project/vllm/issues/1189#issuecomment-1738255066 |
openrouter-runner | github_2023 | python | 12 | OpenRouterTeam | alexanderatallah | @@ -8,6 +8,7 @@ def _to_lower_list(l: List[str]):
vllm_13b_model_ids = [
+ "mistralai/Mistral-7B-Instruct-v0.1", | do we need the same hardware for this model? (or if we stick with 2 gpus, will they scale efficiently with a 7b?) |
openrouter-runner | github_2023 | others | 8 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,32 @@
+# Add new model
+
+1. Copy one of the model file in `aux/models` and rename it to your model name.
+2. Change the `model_id` to the HF model ID.
+3. Adapt the Model class name accordingly.
+4. Import the model into [](./models/__init__.py) and add it to the get_model function.
+
+# Configuration
+
+Se... | you mean `aux/main.py`? |
openrouter-runner | github_2023 | python | 8 | OpenRouterTeam | alexanderatallah | @@ -1,4 +1,18 @@
-from modal.config import Config
+from modal.config import Config as ModalConfig
+from pydantic import BaseModel
-modal_config = Config()
+modal_config = ModalConfig()
keep_warm = None if modal_config.get(key="environment") == "dev" else 1 | shouldn't we always cold-start now? |
openrouter-runner | github_2023 | python | 8 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,14 @@
+from modal import Stub
+from vllm_runner.shared.config import Config
+
+config = Config(
+ name="aux",
+ api_key_id="AUX_API_KEY",
+ download_dir="/model",
+ num_gpu=1,
+ max_batched_tokens=4096,
+ idle_timeout=5 * 60, # 5 minutes | Small comment: this is slightly confusing because `Config` is coming out of `vllm_runner`, which makes me think it's vllm-specific, but it's also being used as the `container_idle_timeout` on Modal containers.
Should we rename vllm_runner to runner, perhaps? |
openrouter-runner | github_2023 | python | 8 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,147 @@
+from vllm_runner.shared.protocol import (
+ Payload,
+ CompletionResponse,
+ ErrorPayload,
+ ErrorResponse,
+)
+from vllm_runner.shared.sampling_params import SamplingParams
+
+from typing import List
+from modal import Secret, method, gpu, Image
+
+from ..shared.common import stub, confi... | does it work to call helper functions from inside each of these class methods? |
openrouter-runner | github_2023 | python | 8 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,147 @@
+from vllm_runner.shared.protocol import (
+ Payload,
+ CompletionResponse,
+ ErrorPayload,
+ ErrorResponse,
+)
+from vllm_runner.shared.sampling_params import SamplingParams
+
+from typing import List
+from modal import Secret, method, gpu, Image
+
+from ..shared.common import stub, confi... | why not use mk_gpu_image? |
openrouter-runner | github_2023 | python | 8 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,201 @@
+# Copied from: https://github.com/vllm-project/vllm/blob/bc0644574ca12d754a031596bdcfe8e1f0e6ab39/vllm/sampling_params.py | can we put something like `# BEGIN OPENROUTER MODIFICATION` / `# END OPENROUTER MODIFICATION` in the code below so we know why we forked it? |
openrouter-runner | github_2023 | python | 8 | OpenRouterTeam | alexanderatallah | @@ -1,5 +1,7 @@
# Copied from: https://github.com/vllm-project/vllm/blob/bc0644574ca12d754a031596bdcfe8e1f0e6ab39/vllm/sampling_params.py
+# The main reason we copy it here is to keep our CPU endpoint container lean - since all we need from vllm is this params validator and not the whole package | thanks 👍 |
openrouter-runner | github_2023 | python | 10 | OpenRouterTeam | louisgv | @@ -124,7 +121,7 @@ def download_model_to_folder():
@stub.cls(
- gpu=gpu.A100(count=NUM_GPU, memory=20),
+ gpu=gpu.A10G(count=NUM_GPU),
secret=Secret.from_name("huggingface"),
allow_concurrent_inputs=12,
container_idle_timeout=600, | Maybe we can reduce the idle timeout as well? |
openrouter-runner | github_2023 | python | 5 | OpenRouterTeam | alexanderatallah | @@ -88,63 +112,114 @@ def download_model_to_folder():
)
)
-stub = Stub("mythalion", image=image)
+stub = Stub(NAME, image=image)
@stub.cls(
- gpu=gpu.A100(),
+ gpu=gpu.A100(count=NUM_GPU, memory=20),
secret=Secret.from_name("huggingface"),
allow_concurrent_inputs=12,
container_idle_tim... | should we make keep_warm 0 when in the dev environment? |
openrouter-runner | github_2023 | typescript | 5 | OpenRouterTeam | alexanderatallah | @@ -2,7 +2,10 @@ import { once } from 'events';
import fs from 'fs/promises';
import { createInterface } from 'readline';
-export async function getWords(n: number, fileNo = 0): Promise<string> {
+export async function getWords( | nit: from the function naming, it was unclear if this fn generates random words or if it gets words out of a file. (had to read it to see that fileNo is actually important and it's doing the latter) |
openrouter-runner | github_2023 | python | 5 | OpenRouterTeam | alexanderatallah | @@ -158,37 +233,46 @@ def completion(
payload: Payload,
token: HTTPAuthorizationCredentials = Depends(auth_scheme),
):
- if token.credentials != os.environ["MYTHALION_API_KEY"]:
+ if token.credentials != os.environ[API_KEY_ID]:
raise HTTPException(
status_code=status.HTTP_401_UNAU... | why do these two things remotely? and if each can take more than 50ms, is it faster to parallelize? |
openrouter-runner | github_2023 | python | 5 | OpenRouterTeam | alexanderatallah | @@ -88,63 +112,114 @@ def download_model_to_folder():
)
)
-stub = Stub("mythalion", image=image)
+stub = Stub(NAME, image=image)
@stub.cls(
- gpu=gpu.A100(),
+ gpu=gpu.A100(count=NUM_GPU, memory=20),
secret=Secret.from_name("huggingface"),
allow_concurrent_inputs=12,
container_idle_tim... | nit: would be nice to add a comment here to explain why we are skipping this |
openrouter-runner | github_2023 | python | 2 | OpenRouterTeam | alexanderatallah | @@ -108,6 +108,7 @@ def __enter__(self):
tensor_parallel_size=1,
# using 95% of GPU memory by default
gpu_memory_utilization=0.95,
+ log_requests=False, | can we undeploy this? see https://modal.com/logs/ap-oeR9RaoFC1krDiG94H2iUq?functionId=fu-W8IPkamMB6habWTtkQRiKJ&taskId=ta-3GWMAUS1dPKhkbTYAZVnh0&inputId=in-pw3u7DxpqrC3ynnn9dstHW |
openrouter-runner | github_2023 | python | 1 | OpenRouterTeam | alexanderatallah | @@ -48,13 +48,14 @@ def download_model_to_folder():
)
)
-stub = Stub("vllm-serving", image=image)
+stub = Stub("mythomax", image=image)
+
@stub.cls(
- gpu=gpu.A100(),
- secret=Secret.from_name("huggingface"),
- allow_concurrent_inputs=12,
- container_idle_timeout=300
+ gpu=gpu.A100(),
+ secret=S... | nit: we estimated around 20 concurrent inputs earlier. what are the cons of increasing this number? |
openrouter-runner | github_2023 | python | 1 | OpenRouterTeam | alexanderatallah | @@ -48,13 +48,14 @@ def download_model_to_folder():
)
)
-stub = Stub("vllm-serving", image=image)
+stub = Stub("mythomax", image=image)
+
@stub.cls(
- gpu=gpu.A100(),
- secret=Secret.from_name("huggingface"),
- allow_concurrent_inputs=12,
- container_idle_timeout=300
+ gpu=gpu.A100(),
+ secret=S... | what does this mean? is it just for the cpu container? how fast is it to cold-start? Not sure we should shut it down after 5 min if it takes 30 sec to cold-start |
openrouter-runner | github_2023 | python | 1 | OpenRouterTeam | alexanderatallah | @@ -112,10 +109,8 @@ async def generate(self, question: str):
print(f"Request completed: {throughput:.4f} tokens/s")
# print(request_output.outputs[0].text)
-@stub.function(
- timeout=60 * 10,
- allow_concurrent_inputs=12
-)
+
+@stub.function(timeout=60 * 10, allow_concurrent_inputs=12)
@web_end... | can we try using https://vllm.readthedocs.io/en/latest/getting_started/quickstart.html#openai-compatible-server ? |
openrouter-runner | github_2023 | python | 1 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,168 @@
+import os
+
+from typing import List, Optional, Union
+
+from fastapi import Depends, HTTPException, status, Request
+
+from modal import Image, Secret, Stub, method, gpu, web_endpoint
+from pydantic import BaseModel
+from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
+
+
+MODEL_DI... | is this where we should specify default settings, or should we do that in the adapter? seems like a good thing for the adapter to own. (for that, Pygmalion has suggested defaults here: https://pygmalionai.github.io/blog/posts/introducing_pygmalion_2/#sillytavern) |
openrouter-runner | github_2023 | python | 1 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,168 @@
+import os
+
+from typing import List, Optional, Union
+
+from fastapi import Depends, HTTPException, status, Request
+
+from modal import Image, Secret, Stub, method, gpu, web_endpoint
+from pydantic import BaseModel
+from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
+
+
+MODEL_DI... | template doesn't seem defined (but prob should be managed by the adapter anyway) |
openrouter-runner | github_2023 | python | 1 | OpenRouterTeam | alexanderatallah | @@ -0,0 +1,168 @@
+import os
+
+from typing import List, Optional, Union
+
+from fastapi import Depends, HTTPException, status, Request
+
+from modal import Image, Secret, Stub, method, gpu, web_endpoint
+from pydantic import BaseModel
+from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
+
+
+MODEL_DI... | skipping this for review, assuming WIP |
dotnet-cloud-native-build-2023 | github_2023 | csharp | 5 | bradygaster | davidfowl | @@ -30,74 +30,80 @@ protected override async Task ExecuteAsync(CancellationToken stoppingToken)
var orders = await _ordersClient.GetOrders();
activity?.AddTag("order-count", orders?.Count ?? 0);
- var orderTasks = new List<Task>();
fore... | These lowercase methods are java |
axum-htmx | github_2023 | others | 15 | robertwayne | robertwayne | @@ -0,0 +1,123 @@
+use axum_core::response::{IntoResponseParts, ResponseParts};
+use http::header::VARY;
+
+use crate::{extractors, headers, HxError};
+
+/// The `Vary: HX-Request` header.
+///
+/// You may want to add this header to the response if your handler responds differently based on
+/// the `HX-Request` reque... | We can use `http::HeaderValue::from_static(HX_TARGET)` instead. This applies to the other instances of `try_into()?` as well.
Originally I was going to say we could make this `type Error = Infallible`, but after thinking some more, `.insert()` can actually panic, so this should probably be addressed in the existing ... |
claude-to-chatgpt | github_2023 | python | 9 | jtsang4 | jtsang4 | @@ -74,15 +74,9 @@ def claude_to_chatgpt_response_stream(self, claude_response, prev_decoded_respon
"object": "chat.completion.chunk",
"created": int(time.time()),
"model": "gpt-3.5-turbo-0301",
- "usage": {
- "prompt_tokens": 0,
- "complet... | The "role" field appeared in my previouse tests, please check it again. |
online-edit-web | github_2023 | typescript | 123 | xun082 | xun082 | @@ -55,6 +61,46 @@ export default function CodeEditor({ editorId }: CodeEditorProps) {
border: isOver ? '1px #3b82f6 solid' : undefined,
};
+ _editor &&
+ _editor.addCommand(monaco.KeyMod.CtrlCmd | monaco.KeyCode.KeyS, () => {
+ const prettierValue = getPrettierConfig(fileData);
+ formatWithPret... | 这里的ts类型别用object，很容易有报错，可以用Record这种 record<string,any>这样，又或者record<string,string|boolean|number> |
online-edit-web | github_2023 | typescript | 119 | xun082 | xun082 | @@ -210,14 +211,23 @@ const SearchPage: FC = () => {
<IncludeAndExclude />
</div>
<div className="mx-7 text-[12px] select-none text-gray-500">
- {searchInpVal ? (
- searchResult && searchResult.length ? (
- <div>{`${resultCount.fileCount} 文件中 ${resultCount.matchCount} 个... | 这里返回null |
create-neat | github_2023 | typescript | 309 | xun082 | uaenaTzx | @@ -146,9 +146,9 @@ export default async function createAppTest(projectName: string, options: Record
if (gitCheck(rootDirectory)) exec("git init", { cwd: rootDirectory });
// 安装传入的依赖
- if (process.env.NODE_ENV === "PROD") {
- await dependenciesInstall(rootDirectory, packageManager);
- }
+ // if (process.e... | 这块为什么要删除
其他插件可能会带进来一些依赖到 pka 里面 |
create-neat | github_2023 | others | 309 | xun082 | uaenaTzx | @@ -0,0 +1,47 @@
+import fs from 'fs';
+import path from 'path';
+import spawn from 'cross-spawn';
+import {fileURLToPath} from 'url'; | 看着没有格式化 |
create-neat | github_2023 | others | 309 | xun082 | uaenaTzx | @@ -0,0 +1,47 @@
+import fs from 'fs';
+import path from 'path';
+import spawn from 'cross-spawn';
+import {fileURLToPath} from 'url';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
+//插件目录
+const pluginsDir = path.resolve(__dirname, '../packages/@plugin');
+
+funct... | log 可以带个颜色（chalk） |
create-neat | github_2023 | others | 309 | xun082 | uaenaTzx | @@ -0,0 +1,47 @@
+import fs from 'fs';
+import path from 'path';
+import spawn from 'cross-spawn';
+import {fileURLToPath} from 'url';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
+//插件目录
+const pluginsDir = path.resolve(__dirname, '../packages/@plugin');
+
+funct... | 看着都没有格式化 |
create-neat | github_2023 | others | 309 | xun082 | uaenaTzx | @@ -0,0 +1,47 @@
+import fs from 'fs';
+import path from 'path';
+import spawn from 'cross-spawn';
+import {fileURLToPath} from 'url';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
+//插件目录
+const pluginsDir = path.resolve(__dirname, '../packages/@plugin');
+
+funct... | 输出信息尽量完整，code 也可以输出 |
create-neat | github_2023 | others | 309 | xun082 | uaenaTzx | @@ -0,0 +1,47 @@
+import fs from 'fs'; | 还缺少一些东西，建议修改的点：
1. 发包过程改成交互式的（参考 cn create）
2. 交互式的具体流程可以包括：是否要批量发，检查当前包是否有更新（否则不用发），是否选择发 beta 包 等等，你再设计看看 |
create-neat | github_2023 | others | 308 | xun082 | uaenaTzx | @@ -72,7 +72,7 @@
},
"engines": {
"node": ">=18",
- "pnpm": "9.4.0"
+ "pnpm": "^9.0.0" | 不要轻易改依赖版本，这个是有什么背景吗
|
create-neat | github_2023 | others | 308 | xun082 | uaenaTzx | @@ -1,5 +1,21 @@
// const path = require("path");
module.exports = {
- rules: [],
+ rules: [ | 这块其实还有很多优化空间，可以看到历史的 config 配置里存在 构建工具 x 框架 的情况，我们看看能不能通过协议的方式来解决问题
如果遇到特殊场景不好做的，可以找我讨论 |
create-neat | github_2023 | javascript | 308 | xun082 | uaenaTzx | @@ -2,11 +2,18 @@ module.exports = {
rules: [
{
test: /\.(ts|tsx|js|jsx)$/,
- exclude: /node_modules/,
- use: {
- loader: "swc-loader",
- exclude: /node_modules/,
- },
+ exclude: [/node_modules/, /public/, /(.|_)min\.js$/], | 同 babel |
create-neat | github_2023 | others | 308 | xun082 | uaenaTzx | @@ -1,7 +1,7 @@
-const PluginConfig = require("./config/index.cjs");
+// const PluginConfig = require("./config/index.cjs");
-const pluginBabel = (buildTool, template) => { | 不需要的就删了吧 |
create-neat | github_2023 | typescript | 308 | xun082 | uaenaTzx | @@ -163,11 +163,12 @@ class Generator {
}
// 根据环境变量加载 plugin/template
+ // 返回增加可选的buildTool，编译器🥱插件(babel/swc)需要 | 这个注释改一下 |
create-neat | github_2023 | typescript | 310 | xun082 | uaenaTzx | @@ -3,21 +3,23 @@ import generator from "@babel/generator";
import fs from "fs-extra";
import chalk from "chalk";
import { parse } from "@babel/parser";
-
-import { relativePathToRoot } from "../utils/constants";
-import { createFiles } from "../utils/createFiles";
-import { createConfigByParseAst } from "../utils/a... | 应该是 ts 后缀 |
create-neat | github_2023 | typescript | 310 | xun082 | uaenaTzx | @@ -231,10 +234,9 @@ class Generator {
// 单独处理一个框架相关依赖，主要是将框架相关的依赖包插入到pkg内，以及将需要的构建工具配置合并到构建工具模板中
async templateGenerate() {
- const templateGenerator = await this.loadBase(
- `packages/core/template/template-${this.templateName}/generator/index.cjs`,
- "",
- );
+ const templatePath = `packag... | log 删了吧 |
create-neat | github_2023 | typescript | 307 | xun082 | uaenaTzx | @@ -18,6 +17,8 @@ import dependenciesInstall from "./dependenciesInstall";
import { createReadmeString } from "./createFiles";
import { buildToolConfigDevDependencies, buildToolScripts } from "./constants";
+const { resolveApp } = require("@laconic/utils"); | 这个文件并没有引用这个呀 |
create-neat | github_2023 | typescript | 306 | xun082 | uaenaTzx | @@ -1,28 +1,35 @@
+/** 通用协议常量，定义了跨不同工具和框架交互的协议类型。*/
const globalProtocol = {
- ENTRY_FILE: "ENTRY_FILE", // 入口文件配置，如：入口文件引入全局 less、scss
+ /** 入口文件配置，例如引入全局的less、scss文件。 */
+ ENTRY_FILE: "ENTRY_FILE",
+ /** 更改文件导出的内容。 */
UPDATE_EXPORT_CONTENT_PROTOCOL: "UPDATE_EXPORT_CONTENT_PROTOCOL",
+ /** 向文件插入import语句。 */
... | 扩展了通用协议，并添加处理样式的协议
这个没必要注释 |
create-neat | github_2023 | typescript | 306 | xun082 | uaenaTzx | @@ -1,28 +1,35 @@
+/** 通用协议常量，定义了跨不同工具和框架交互的协议类型。*/
const globalProtocol = {
- ENTRY_FILE: "ENTRY_FILE", // 入口文件配置，如：入口文件引入全局 less、scss
+ /** 入口文件配置，例如引入全局的less、scss文件。 */
+ ENTRY_FILE: "ENTRY_FILE",
+ /** 更改文件导出的内容。 */
UPDATE_EXPORT_CONTENT_PROTOCOL: "UPDATE_EXPORT_CONTENT_PROTOCOL",
+ /** 向文件插入import语句。 */
... | 根据框架，不同的打包工具需要不同的插件，有些是都需要用的，有些是框架独有的
解释有些奇怪，add config 不一定是添加插件配置，描述有些局限 |
create-neat | github_2023 | typescript | 306 | xun082 | uaenaTzx | @@ -1,28 +1,35 @@
+/** 通用协议常量，定义了跨不同工具和框架交互的协议类型。*/
const globalProtocol = {
- ENTRY_FILE: "ENTRY_FILE", // 入口文件配置，如：入口文件引入全局 less、scss
+ /** 入口文件配置，例如引入全局的less、scss文件。 */
+ ENTRY_FILE: "ENTRY_FILE",
+ /** 更改文件导出的内容。 */
UPDATE_EXPORT_CONTENT_PROTOCOL: "UPDATE_EXPORT_CONTENT_PROTOCOL",
+ /** 向文件插入import语句。 */
... | slot 是什么也得说明清楚，比如：对目标文件的指定插槽进行内容注入 |
create-neat | github_2023 | typescript | 306 | xun082 | uaenaTzx | @@ -1,28 +1,35 @@
+/** 通用协议常量，定义了跨不同工具和框架交互的协议类型。*/
const globalProtocol = {
- ENTRY_FILE: "ENTRY_FILE", // 入口文件配置，如：入口文件引入全局 less、scss
+ /** 入口文件配置，例如引入全局的less、scss文件。 */ | 英文字母和中文留空格，比如 xx 这样 |
create-neat | github_2023 | typescript | 298 | xun082 | uaenaTzx | @@ -66,6 +66,11 @@ export const buildToolScripts = {
},
};
+export const templateFileExtension = { | 注释明确 |
create-neat | github_2023 | typescript | 298 | xun082 | uaenaTzx | @@ -174,13 +197,18 @@ class FileTree {
* @param url 添加文件的原始的真实路径
* @param parentDir 父文件夹路径
*/
- addToTreeByTemplateDirPathAndEjs(url: string, parentDir: string, options: any) {
+ addToTreeByTemplateDirPathAndEjs(url: string, parentDir: string, options: any, isTs: any) { | any 不对，而且这里是可以读取到 process 的值的，不太需要传参 |
create-neat | github_2023 | typescript | 298 | xun082 | uaenaTzx | @@ -107,13 +108,25 @@ class FileTree {
return file;
}
+ private transformFileExtension(fileContent: string, fileName: string, fileExtension: string) { | 注释尽量补充完善 |
create-neat | github_2023 | others | 302 | xun082 | uaenaTzx | @@ -1,16 +1,104 @@
-module.exports = (generatorAPI) => {
+const path = require("path");
+const fs = require("fs");
+
+function fileRender(files) {
+ try {
+ const outputDir = path.join(__dirname, "template");
+ Object.entries(files).forEach(([filePath, content]) => {
+ const fullPath = path.join(outputDir, ... | 这个不走 template 渲染吗 |
create-neat | github_2023 | others | 302 | xun082 | uaenaTzx | @@ -1,16 +1,104 @@
-module.exports = (generatorAPI) => {
+const path = require("path");
+const fs = require("fs");
+
+function fileRender(files) {
+ try {
+ const outputDir = path.join(__dirname, "template");
+ Object.entries(files).forEach(([filePath, content]) => {
+ const fullPath = path.join(outputDir, ... | 包管理器不是写死的 |
create-neat | github_2023 | others | 302 | xun082 | uaenaTzx | @@ -1,16 +1,104 @@
-module.exports = (generatorAPI) => {
+const path = require("path");
+const fs = require("fs");
+
+function fileRender(files) {
+ try {
+ const outputDir = path.join(__dirname, "template");
+ Object.entries(files).forEach(([filePath, content]) => {
+ const fullPath = path.join(outputDir, ... | +1 |
create-neat | github_2023 | typescript | 302 | xun082 | uaenaTzx | @@ -8,6 +8,8 @@ const globalProtocol = {
const pluginToTemplateProtocol = {
...globalProtocol,
PROCESS_STYLE_PLUGIN: "PROCESS_STYLE_PLUGIN",
+ INSERT_IMPORT: "INSERT_IMPORT",
+ RENDER_FILE: "RENDER_FILE", | 没看到相关的处理器，不过这个协议确实意义不大 |
create-neat | github_2023 | others | 299 | xun082 | uaenaTzx | @@ -35,11 +35,22 @@ module.exports = (generatorAPI) => {
[pluginToTemplateProtocol.UPDATE_EXPORT_CONTENT_PROTOCOL]: {
params: {
url: 'src/App',
- exportContent: 'Observer',
+ exportContent: 'observer', | 这个改动是有什么原因 |
create-neat | github_2023 | others | 299 | xun082 | uaenaTzx | @@ -35,11 +35,22 @@ module.exports = (generatorAPI) => {
[pluginToTemplateProtocol.UPDATE_EXPORT_CONTENT_PROTOCOL]: {
params: {
url: 'src/App',
- exportContent: 'Observer',
+ exportContent: 'observer',
astOptions: {
parserOptions: { sourceType: "module", plugins: [... | 内容修正 |
create-neat | github_2023 | javascript | 299 | xun082 | uaenaTzx | @@ -2,7 +2,9 @@ import React from "react";
import "./index.css";
function App() {
+ // <!--slot: store-slot --> | 不用加 // 容易出现歧义 |
create-neat | github_2023 | typescript | 299 | xun082 | uaenaTzx | @@ -0,0 +1,24 @@
+import prettier from "prettier";
+import prettierPluginVue from "prettier-plugin-vue"; | 只有 vue 的吗，这个不能搞成通用的吗 |
create-neat | github_2023 | others | 295 | xun082 | uaenaTzx | @@ -10,9 +10,31 @@ module.exports = (generatorAPI) => {
});
generatorAPI.protocolGenerate({
+ [pluginToTemplateProtocol.INSERT_IMPORT_PROTOCOL]: {
+ params: {
+ imports: [
+ {
+ dir: "src/App",
+ module: [ | modules吧 |
create-neat | github_2023 | others | 295 | xun082 | uaenaTzx | @@ -10,9 +10,31 @@ module.exports = (generatorAPI) => {
});
generatorAPI.protocolGenerate({
+ [pluginToTemplateProtocol.INSERT_IMPORT_PROTOCOL]: {
+ params: {
+ imports: [
+ {
+ dir: "src/App",
+ module: [
+ {
+ name: "mobx",
+ ... | 这块的传参确实有一定学习成本 |
create-neat | github_2023 | typescript | 294 | xun082 | uaenaTzx | @@ -0,0 +1,35 @@
+import path from "path";
+
+import { FileData } from "../models/FileTree";
+
+/**
+ * 根据给定的文件路径，从嵌套的文件结构中检索目标文件的数据。
+ * | jsdoc 不需要这一行 |
create-neat | github_2023 | typescript | 294 | xun082 | uaenaTzx | @@ -52,3 +52,14 @@ export const createCallExpression = (
*/
export const createObjectProperty = (name: string, value: Parameters<typeof objectProperty>[1]) =>
objectProperty(identifier(name), value);
+
+/**
+ * 封装遍历ast中ExportDefaultDeclaration的通用函数
+ * @param {object} path
+ * @param {object} t
+ * @param {string... | any 尽量避免吧 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.