before_code stringlengths 30 149k | reviewer_comment stringlengths 16 20.6k | after_code stringlengths 30 163k | diff_context stringlengths 0 22.4k | file_path stringlengths 6 155 | comment_line int64 0 26 | language stringclasses 19
values | quality_score float64 0.07 1 | comment_type stringclasses 9
values | comment_length int64 16 20.6k | before_lines int64 2 4.61k | after_lines int64 2 4.4k | is_negative bool 2
classes | pr_title stringlengths 7 140 | pr_number int64 18 61.2k | repo_name stringclasses 29
values | repo_stars int64 321 56k | repo_language stringclasses 14
values | reviewer_username stringclasses 378
values | author_username stringclasses 739
values | user stringlengths 361 200k | assistant stringlengths 16 20.6k | system stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
import axios, { AxiosRequestConfig } from "axios";
import type { JWT } from "next-auth/jwt";
import { ChatItem, InferenceMessage, InferencePostMessageResponse } from "src/types/Chat";
import type { Readable } from "stream";
export class OasstInferenceClient {
// this is not a long lived class, this is why the token ... | NEXT_PUBLIC_INFERENCE_SERVER_HOST will be undefined in production. We need to pass it from getServerSideProps. Or we can use getInitialProps in the `_app.tsx` file and pass it to a context or any global variable. | import axios, { AxiosRequestConfig } from "axios";
import { JWT } from "next-auth/jwt";
import {
ChatItem,
InferenceMessage,
InferencePostMessageResponse,
ModelInfo,
WorkParametersInput,
} from "src/types/Chat";
import type { Readable } from "stream";
export class OasstInferenceClient {
// this is not a lo... | @@ -1,69 +1,38 @@
-/* eslint-disable @typescript-eslint/no-explicit-any */
import axios, { AxiosRequestConfig } from "axios";
-import Cookies from "cookies";
-import type { NextApiRequest, NextApiResponse } from "next";
-import { JWT } from "next-auth/jwt";
-import { ChatItem, InferenceTokenResponse, InferenceMessage,... | website/src/lib/oasst_inference_client.ts | 11 | TypeScript | 0.786 | suggestion | 212 | 36 | 36 | false | Login to inference through website | 2,176 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/lib/oasst_inference_client.ts
PR Title: Login to inference through website
BEFORE CODE: import axios, { AxiosRequestConfig } from "axios";
import type { JWT } from "next-auth/jwt";
import { ChatItem, InferenceMessage, InferencePostMessageRespon... | NEXT_PUBLIC_INFERENCE_SERVER_HOST will be undefined in production. We need to pass it from getServerSideProps. Or we can use getInitialProps in the `_app.tsx` file and pass it to a context or any global variable. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
self,
pooling: Literal["mean", "last"] = "last",
**kwargs,
):
super().__init__(**kwargs)
self.pooling = pooling or "last"
@dataclass
class GPTNeoXRewardModelOutput(ModelOutput):
"""
Reward model output.
Args:
logits (`torch.FloatTensor` of shape `(batch... | I personally prefer `isinstance(config, GPTNeoXConfig)` | self,
pooling: Literal["mean", "last"] = "last",
**kwargs,
):
super().__init__(**kwargs)
self.pooling = pooling or "last"
@dataclass
class GPTNeoXRewardModelOutput(ModelOutput):
"""
Reward model output.
Args:
logits (`torch.FloatTensor` of shape `(batch... | @@ -1,68 +1,94 @@
-import os
-from typing import Literal, Optional, Union
+from dataclasses import dataclass
+from typing import Literal, Optional
+import torch
import torch.nn as nn
-from models import get_specific_model
-from transformers import PretrainedConfig, PreTrainedModel
+from transformers import AutoConfi... | model/model_training/models/reward_model.py | 26 | Python | 0.429 | suggestion | 55 | 51 | 51 | false | Add GPTNeoXRewardModel | 2,182 | LAION-AI/Open-Assistant | 29,137 | Python | dvruette | andreaskoepf | Review this pull request change.
Language: Python
File: model/model_training/models/reward_model.py
PR Title: Add GPTNeoXRewardModel
BEFORE CODE: self,
pooling: Literal["mean", "last"] = "last",
**kwargs,
):
super().__init__(**kwargs)
self.pooling = pooling or "last"
@d... | I personally prefer `isinstance(config, GPTNeoXConfig)` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import transformers
import utils
import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import inference
from settings import settings
def terminate_worker(signum, frame):
logger.warning(f"Signal {signum}. Terminating worker...")
sys.exit(0)
d... | Minor but we could avoid the nesting here | import transformers
import utils
import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import inference
from settings import settings
def terminate_worker(signum, frame):
logger.warning(f"Signal {signum}. Terminating worker...")
sys.exit(0)
d... | @@ -23,19 +24,37 @@ def main():
signal.signal(signal.SIGINT, terminate_worker)
logger.info(f"Inference protocol version: {inference.INFERENCE_PROTOCOL_VERSION}")
- if settings.model_id != "_lorem":
- if "llama" in settings.model_id:
- tokenizer: transformers.PreTrainedTokenizer = transf... | inference/worker/__main__.py | 26 | Python | 0.143 | nitpick | 41 | 51 | 51 | false | Introduce model configs to abstract pairings of models and hardware | 2,194 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | yk | Review this pull request change.
Language: Python
File: inference/worker/__main__.py
PR Title: Introduce model configs to abstract pairings of models and hardware
BEFORE CODE: import transformers
import utils
import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_sh... | Minor but we could avoid the nesting here | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import utils
import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import inference
from settings import settings
def terminate_worker(signum, frame):
logger.warning(f"Signal {signum}. Terminating worker...")
sys.exit(0)
def main():
signa... | is this special handling still necessary with latest transformers version? (e.g. maybe wee need to update our json config files?) | import utils
import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import inference
from settings import settings
def terminate_worker(signum, frame):
logger.warning(f"Signal {signum}. Terminating worker...")
sys.exit(0)
def main():
signa... | @@ -23,19 +24,37 @@ def main():
signal.signal(signal.SIGINT, terminate_worker)
logger.info(f"Inference protocol version: {inference.INFERENCE_PROTOCOL_VERSION}")
- if settings.model_id != "_lorem":
- if "llama" in settings.model_id:
- tokenizer: transformers.PreTrainedTokenizer = transf... | inference/worker/__main__.py | 26 | Python | 0.357 | question | 129 | 51 | 51 | false | Introduce model configs to abstract pairings of models and hardware | 2,194 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | yk | Review this pull request change.
Language: Python
File: inference/worker/__main__.py
PR Title: Introduce model configs to abstract pairings of models and hardware
BEFORE CODE: import utils
import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import ... | is this special handling still necessary with latest transformers version? (e.g. maybe wee need to update our json config files?) | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import sys
from oasst_shared import model_configs
from settings import settings
if __name__ == "__main__":
model_config = model_configs.MODEL_CONFIGS.get(settings.model_config_name)
if model_config is None:
print(f"Unknown model config name: {settings.model_config_name}")
sys.exit(2)
prop ... | maybe add `len(sys.argv) > 1` check (since everything else is here handled gracefully) | import sys
from oasst_shared import model_configs
from settings import settings
if __name__ == "__main__":
model_config = model_configs.MODEL_CONFIGS.get(settings.model_config_name)
if model_config is None:
print(f"Unknown model config name: {settings.model_config_name}")
sys.exit(2)
if le... | @@ -0,0 +1,18 @@
+import sys
+
+from oasst_shared import model_configs
+from settings import settings
+
+if __name__ == "__main__":
+ model_config = model_configs.MODEL_CONFIGS.get(settings.model_config_name)
+ if model_config is None:
+ print(f"Unknown model config name: {settings.model_config_name}")
+ ... | inference/worker/get_model_config_prop.py | 14 | Python | 0.5 | suggestion | 86 | 19 | 22 | false | Introduce model configs to abstract pairings of models and hardware | 2,194 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | yk | Review this pull request change.
Language: Python
File: inference/worker/get_model_config_prop.py
PR Title: Introduce model configs to abstract pairings of models and hardware
BEFORE CODE: import sys
from oasst_shared import model_configs
from settings import settings
if __name__ == "__main__":
model_config =... | maybe add `len(sys.argv) > 1` check (since everything else is here handled gracefully) | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
const [messages, setMessages] = useState<InferenceMessage[]>([]);
const [streamedResponse, setResponse] = useState<string | null>(null);
const [queueInfo, setQueueInfo] = useState<QueueInfo | null>(null);
const [isSending, setIsSending] = useBoolean();
useSWR<ChatItem>(API_ROUTES.GET_CHAT(chatId), get, {
... | Have you tested the top_k here? Andreas suggest an exponential function for this. |
const [messages, setMessages] = useState<InferenceMessage[]>([]);
const [streamedResponse, setResponse] = useState<string | null>(null);
const [queueInfo, setQueueInfo] = useState<QueueInfo | null>(null);
const [isSending, setIsSending] = useBoolean();
useSWR<ChatItem>(API_ROUTES.GET_CHAT(chatId), get, {
... | @@ -29,35 +35,34 @@ export const ChatConversation = ({ chatId }: ChatConversationProps) => {
const [queueInfo, setQueueInfo] = useState<QueueInfo | null>(null);
const [isSending, setIsSending] = useBoolean();
- useSWR<ChatItem>(API_ROUTES.GET_CHAT(chatId), get, { onSuccess: (chat) => setMessages(chat.messages)... | website/src/components/Chat/ChatConversation.tsx | 26 | TypeScript | 0.214 | question | 81 | 51 | 51 | false | Use new inference model config / API | 2,208 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConversation.tsx
PR Title: Use new inference model config / API
BEFORE CODE:
const [messages, setMessages] = useState<InferenceMessage[]>([]);
const [streamedResponse, setResponse] = useState<string | null>(null);
con... | Have you tested the top_k here? Andreas suggest an exponential function for this. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
class TokenResponse(WorkerResponseBase):
response_type: Literal["token"] = "token"
text: str
log_prob: float | None
token_id: int
class GeneratedTextResponse(WorkerResponseBase):
response_type: Literal["generated_text"] = "generated_text"
text: str
finish_reason: Literal["length", "eos_to... | this belongs to a PR stacked on top of this one | metrics: WorkerMetricsInfo | None = None
class TokenResponse(WorkerResponseBase):
response_type: Literal["token"] = "token"
text: str
log_prob: float | None
token_id: int
class GeneratedTextResponse(WorkerResponseBase):
response_type: Literal["generated_text"] = "generated_text"
text: st... | @@ -237,17 +239,18 @@ class InternalFinishedMessageResponse(WorkerResponseBase):
class InternalErrorResponse(WorkerResponseBase):
response_type: Literal["internal_error"] = "internal_error"
error: str
+ message: MessageRead | oasst-shared/oasst_shared/schemas/inference.py | 26 | Python | 0.071 | suggestion | 47 | 51 | 51 | false | Create worker metrics manually for more control | 2,229 | LAION-AI/Open-Assistant | 29,137 | Python | yk | yk | Review this pull request change.
Language: Python
File: oasst-shared/oasst_shared/schemas/inference.py
PR Title: Create worker metrics manually for more control
BEFORE CODE:
class TokenResponse(WorkerResponseBase):
response_type: Literal["token"] = "token"
text: str
log_prob: float | None
token_id:... | this belongs to a PR stacked on top of this one | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
await sendVote({ chat_id: chatId, message_id: messageId, score: newScore });
} catch {
// TODO maybe we should trigger a toast or something
// revert on any error
setMessages((messages) =>
messages.map((message) => {
return message.id !== messageId ? message :... | can you add a `useCallback` here? | await sendVote({ chat_id: chatId, message_id: messageId, score: newScore });
} catch {
// TODO maybe we should trigger a toast or something
// revert on any error
setMessages((messages) =>
messages.map((message) => {
return message.id !== messageId ? message :... | @@ -129,11 +185,13 @@ export const ChatConversation = ({ chatId }: ChatConversationProps) => {
score={message.score}
state={message.state}
onVote={handleOnVote}
+ onRetry={() => handleOnRetry(message.parent_id!)} | website/src/components/Chat/ChatConversation.tsx | 26 | TypeScript | 0.286 | question | 33 | 51 | 51 | false | Enabling Threads and Retry for Web Chat | 2,232 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | yk | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConversation.tsx
PR Title: Enabling Threads and Retry for Web Chat
BEFORE CODE: await sendVote({ chat_id: chatId, message_id: messageId, score: newScore });
} catch {
// TODO maybe we should trigger a t... | can you add a `useCallback` here? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
);
const send = useCallback(async () => {
const content = inputRef.current?.value.trim();
if (!content || !chatId) {
return;
}
setIsSending.on();
// find last VALID assistant message
const parent_id =
currentThread
.slice()
.reverse()
.find((m) => m.role... | ```suggestion
await initiate_assistant_message(prompter_message.id);
``` | );
const send = useCallback(async () => {
const content = inputRef.current?.value.trim();
if (!content || !chatId) {
return;
}
setIsSending.on();
// find last VALID assistant message
const parent_id =
currentThread
.slice()
.reverse()
.find((m) => m.role... | @@ -48,53 +124,33 @@ export const ChatConversation = ({ chatId }: ChatConversationProps) => {
// find last VALID assistant message
const parent_id =
- messages
+ currentThread
.slice()
.reverse()
.find((m) => m.role === "assistant" && m.state === "complete")?.id ?? null;... | website/src/components/Chat/ChatConversation.tsx | 26 | TypeScript | 0.571 | suggestion | 77 | 51 | 51 | false | Enabling Threads and Retry for Web Chat | 2,232 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | yk | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConversation.tsx
PR Title: Enabling Threads and Retry for Web Chat
BEFORE CODE: );
const send = useCallback(async () => {
const content = inputRef.current?.value.trim();
if (!content || !chatId) {
return;
... | ```suggestion
await initiate_assistant_message(prompter_message.id);
``` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
raise fastapi.HTTPException(status_code=400, detail="Only assistant messages can be streamed.")
if message.has_finished:
raise fastapi.HTTPException(status_code=204, detail=message.state)
async def event_generator(chat_id: str, message_id: str, worker_compat_hash: str | None):
redis_cl... | could this be done in parallel like this? or is this a race condition?
```python
[qdeq, qenq, mpos] = await asyncio.gather(
work_queue.get_deq_counter(),
work_queue.get_enq_counter(),
queueing.get_pos_value(redis_client, message_id)
)
``` | message: models.DbMessage = await ucr.get_message_by_id(chat_id=chat_id, message_id=message_id)
if message.role != "assistant":
raise fastapi.HTTPException(status_code=400, detail="Only assistant messages can be streamed.")
if message.has_finished:
raise fastapi.HTTPException(status_cod... | @@ -133,19 +134,33 @@ async def message_events(
if message.has_finished:
raise fastapi.HTTPException(status_code=204, detail=message.state)
- async def event_generator(chat_id: str, message_id: str):
+ async def event_generator(chat_id: str, message_id: str, worker_compat_hash: str | None):
... | inference/server/oasst_inference_server/routes/chats.py | 26 | Python | 1 | bug | 253 | 51 | 51 | false | Computing message queue positions | 2,235 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | yk | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/routes/chats.py
PR Title: Computing message queue positions
BEFORE CODE: raise fastapi.HTTPException(status_code=400, detail="Only assistant messages can be streamed.")
if message.has_finished:
raise... | could this be done in parallel like this? or is this a race condition?
```python
[qdeq, qenq, mpos] = await asyncio.gather(
work_queue.get_deq_counter(),
work_queue.get_enq_counter(),
queueing.get_pos_value(redis_client, message_id)
)
``` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
self.tokens.pop()
yield from self.tokens
else:
yield from self.tokens
def wait_for_inference_server(inference_server_url: str, timeout: int = 600):
health_url = f"{inference_server_url}/health"
time_limit = time.time() + timeout
while True:
try:
... | did not understand how this helps in computing message queue positions. is this needed ? | self.tokens.pop()
yield from self.tokens
else:
yield from self.tokens
def wait_for_inference_server(inference_server_url: str, timeout: int = 600):
health_url = f"{inference_server_url}/health"
time_limit = time.time() + timeout
while True:
try:
... | @@ -76,6 +76,7 @@ def wait_for_inference_server(inference_server_url: str, timeout: int = 600):
def text_to_events(text: str, seed: int | None = None, pause: float = 0.0):
+ time.sleep(2) | inference/worker/utils.py | 26 | Python | 0.214 | question | 88 | 51 | 51 | false | Computing message queue positions | 2,235 | LAION-AI/Open-Assistant | 29,137 | Python | melvinebenezer | yk | Review this pull request change.
Language: Python
File: inference/worker/utils.py
PR Title: Computing message queue positions
BEFORE CODE: self.tokens.pop()
yield from self.tokens
else:
yield from self.tokens
def wait_for_inference_server(inference_server_url: str, ... | did not understand how this helps in computing message queue positions. is this needed ? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import redis.asyncio as redis
from oasst_inference_server.settings import settings
class RedisQueue:
def __init__(
self,
redis_client: redis.Redis,
queue_id: str,
expire: int | None = None,
with_counter: bool = False,
counter_pos_expire: int = 1,
) -> None:
... | This is type hinted to return `None` but may not | import redis.asyncio as redis
from oasst_inference_server.settings import settings
class RedisQueue:
def __init__(
self,
redis_client: redis.Redis,
queue_id: str,
expire: int | None = None,
with_counter: bool = False,
counter_pos_expire: int = 1,
) -> None:
... | @@ -3,36 +3,76 @@
class RedisQueue:
- def __init__(self, redis_client: redis.Redis, queue_id: str) -> None:
+ def __init__(
+ self,
+ redis_client: redis.Redis,
+ queue_id: str,
+ expire: int | None = None,
+ with_counter: bool = False,
+ counter_pos_expire: int = 1,... | inference/server/oasst_inference_server/queueing.py | 20 | Python | 0.357 | suggestion | 48 | 45 | 45 | false | Computing message queue positions | 2,235 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | yk | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/queueing.py
PR Title: Computing message queue positions
BEFORE CODE: import redis.asyncio as redis
from oasst_inference_server.settings import settings
class RedisQueue:
def __init__(
self,
redis_cl... | This is type hinted to return `None` but may not | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
"""
Helper function for patching HF language models.
Currently supports: GPTNeoX-based models
Limitations:
- Flash attention requires CUDA and fp16/bf16 training. It also requires contiguous attention masks.
- Residual dropout does not support multi-GPU training without DeepDpeed.
"""
... | `and not resid_pdrop` would also be possible. | Helper function for patching HF language models.
Currently supports: GPTNeoX-based models
Limitations:
- Flash attention requires CUDA and fp16/bf16 training. It also requires contiguous attention masks.
- Residual dropout does not support multi-GPU training without DeepDpeed.
"""
globa... | @@ -127,6 +127,9 @@ def patch_model(
if resid_pdrop is not None and (resid_pdrop < 0 or resid_pdrop > 1.0):
raise ValueError("Invalid argument: `resid_pdrop` must be between 0.0 and 1.0")
+ if not flash_attention and (resid_pdrop is None or resid_pdrop == 0.0): | model/model_training/models/patching.py | 26 | Python | 0.214 | suggestion | 45 | 51 | 51 | false | Fix GPTNeoX-20B training | 2,240 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | dvruette | Review this pull request change.
Language: Python
File: model/model_training/models/patching.py
PR Title: Fix GPTNeoX-20B training
BEFORE CODE: """
Helper function for patching HF language models.
Currently supports: GPTNeoX-based models
Limitations:
- Flash attention requires CUDA and fp16/b... | `and not resid_pdrop` would also be possible. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
type AppPropsWithLayout = AppProps & {
Component: NextPageWithLayout;
env: typeof process.env;
cookie: string;
};
const swrConfig: SWRConfiguration = {
revalidateOnFocus: false,
revalidateOnMount: true,
};
function MyApp({ Component, pageProps: { session, ...pageProps }, env, cookie }: AppPropsWithLayout) {... | could we maybe do it like this?
```ts
process.env = new Proxy(env, {
get(env, key) {
const value = env[key];
if (!value) {
throw new Error("Environment variable not set in _app.tsx")
}
return value;
}
)
```
I think it is very easy to make this mistake, wh... | type AppPropsWithLayout = AppProps & {
Component: NextPageWithLayout;
env: typeof process.env;
cookie: string;
};
const swrConfig: SWRConfiguration = {
revalidateOnFocus: false,
revalidateOnMount: true,
};
function MyApp({ Component, pageProps: { session, ...pageProps }, env, cookie }: AppPropsWithLayout) {... | @@ -1,43 +1,56 @@
import "../styles/globals.css";
import "focus-visible";
-import type { AppProps } from "next/app";
+import { boolean } from "boolean";
+import type { AppContext, AppProps } from "next/app";
import Head from "next/head";
import { SessionProvider } from "next-auth/react";
import { appWithTranslat... | website/src/pages/_app.tsx | 26 | TypeScript | 1 | suggestion | 442 | 51 | 51 | false | Expose env vars globally | 2,244 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/pages/_app.tsx
PR Title: Expose env vars globally
BEFORE CODE: type AppPropsWithLayout = AppProps & {
Component: NextPageWithLayout;
env: typeof process.env;
cookie: string;
};
const swrConfig: SWRConfiguration = {
revalidateOnFocus: f... | could we maybe do it like this?
```ts
process.env = new Proxy(env, {
get(env, key) {
const value = env[key];
if (!value) {
throw new Error("Environment variable not set in _app.tsx")
}
return value;
}
)
```
I think it is very easy to make this mistake, wh... | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import "../styles/globals.css";
import "focus-visible";
import { boolean } from "boolean";
import type { AppContext, AppProps } from "next/app";
import Head from "next/head";
import { SessionProvider } from "next-auth/react";
import { appWithTranslation, useTranslation } from "next-i18next";
import React, { useEffect ... | can you check if this is still needed? I remember that I did some debugging, and this cookie field was always empty. | <>
<Head>
<meta name="description" key="description" content={t("index:description")} />
</Head>
<FlagsProvider value={flags}>
<Chakra cookie={cookie}>
<SWRConfig value={swrConfig}>
<SessionProvider session={session}>{page}</SessionProvider>
</SWRCon... | @@ -46,5 +59,18 @@ function MyApp({ Component, pageProps: { session, cookies, ...pageProps } }: App
</>
);
}
-export { getServerSideProps };
+
+MyApp.getInitialProps = ({ ctx: { req } }: AppContext) => {
+ return {
+ env: {
+ INFERENCE_SERVER_HOST: process.env.INFERENCE_SERVER_HOST,
+ ENABLE_CHAT... | website/src/pages/_app.tsx | 26 | TypeScript | 0.5 | bug | 116 | 77 | 31 | false | Expose env vars globally | 2,244 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/pages/_app.tsx
PR Title: Expose env vars globally
BEFORE CODE: import "../styles/globals.css";
import "focus-visible";
import { boolean } from "boolean";
import type { AppContext, AppProps } from "next/app";
import Head from "next/head";
impor... | can you check if this is still needed? I remember that I did some debugging, and this cookie field was always empty. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
"llama": TokenizerConfig(special_tokens=SpecialTokens("</s>", "</s>", sep_token="<s>")),
"cerebras": TokenizerConfig(special_tokens=SpecialTokens("<|endoftext|>", "<|endoftext|>", "<|endoftext|>")),
}
def match_tokenizer_name(model_name: str) -> TokenizerConfig:
"""
Match a partial model name to a tok... | does this have to be specified? Are the other models released without a tokenizer-config? Otherwise this coud be removed .. the special handling for LLaMA was also removed already... | "gpt-neox": TokenizerConfig(special_tokens=SpecialTokens("<|padding|>", "<|endoftext|>", "<|endoftext|>")),
"llama": TokenizerConfig(special_tokens=SpecialTokens("</s>", "</s>", sep_token="<s>")),
"cerebras": TokenizerConfig(special_tokens=SpecialTokens("<|endoftext|>", "<|endoftext|>", "<|endoftext|>")),
}... | @@ -189,6 +190,9 @@ def get_tokenizer(conf) -> transformers.AutoTokenizer:
# explicitly specify LLaMATokenizer class until AutoTokenizer works
# assumes that the tokenizer config is stored in the same directory as the model weights
tokenizer = LLaMATokenizer.from_pretrained(conf.model_name)
+... | model/model_training/utils.py | 26 | Python | 0.429 | question | 182 | 51 | 51 | false | Add support for Cerebras-GPT for training | 2,276 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | olliestanley | Review this pull request change.
Language: Python
File: model/model_training/utils.py
PR Title: Add support for Cerebras-GPT for training
BEFORE CODE: "llama": TokenizerConfig(special_tokens=SpecialTokens("</s>", "</s>", sep_token="<s>")),
"cerebras": TokenizerConfig(special_tokens=SpecialTokens("<|endoftex... | does this have to be specified? Are the other models released without a tokenizer-config? Otherwise this coud be removed .. the special handling for LLaMA was also removed already... | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import axios, { AxiosRequestConfig } from "axios";
import { JWT } from "next-auth/jwt";
import {
ChatItem,
InferenceMessage,
InferencePostAssistantMessageParams,
InferencePostPrompterMessageParams,
ModelInfo,
TrustedClient,
} from "src/types/Chat";
import type { Readable } from "stream";
export class Oasst... | Is this the correct uuid in the data collection backend? | stream_events({ chat_id, message_id }: { chat_id: string; message_id: string }) {
return this.request<Readable>(`/chats/${chat_id}/messages/${message_id}/events`, {
headers: {
Accept: "text/event-stream",
Connection: "keep-alive",
"Cache-Control": "no-cache, no-transform",
},
... | @@ -81,6 +75,13 @@ export class OasstInferenceClient {
}
export const createInferenceClient = (jwt: JWT) => {
- const token = jwt.inferenceTokens?.access_token.access_token;
+ const info: TrustedClient = {
+ // TODO: use api key from env
+ api_key: "0000",
+ client: "website",
+ user_id: jwt.sub, | website/src/lib/oasst_inference_client.ts | 26 | TypeScript | 0.214 | question | 56 | 88 | 32 | false | Use trusted clients for inference auth | 2,278 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/lib/oasst_inference_client.ts
PR Title: Use trusted clients for inference auth
BEFORE CODE: import axios, { AxiosRequestConfig } from "axios";
import { JWT } from "next-auth/jwt";
import {
ChatItem,
InferenceMessage,
InferencePostAssistan... | Is this the correct uuid in the data collection backend? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
TrustedClient: this.clientToken,
},
});
return data;
}
inference_login() {
return this.request("/auth/trusted", { method: "POST" });
}
get_my_chats(): Promise<ChatItem[]> {
return this.request("/chats");
}
async create_chat(): Promise<ChatItem> {
const create = () => thi... | Can we check for the http status here? | TrustedClient: this.clientToken,
},
});
return data;
}
inference_login() {
return this.request("/auth/trusted", { method: "POST" });
}
get_my_chats(): Promise<ChatItem[]> {
return this.request("/chats");
}
async create_chat(): Promise<ChatItem> {
const create = () => thi... | @@ -3,45 +3,49 @@ import { JWT } from "next-auth/jwt";
import {
ChatItem,
InferenceMessage,
- InferencePostPrompterMessageParams,
InferencePostAssistantMessageParams,
+ InferencePostPrompterMessageParams,
ModelInfo,
+ TrustedClient,
} from "src/types/Chat";
import type { Readable } from "stream";
e... | website/src/lib/oasst_inference_client.ts | 26 | TypeScript | 0.071 | question | 38 | 51 | 51 | false | Use trusted clients for inference auth | 2,278 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/lib/oasst_inference_client.ts
PR Title: Use trusted clients for inference auth
BEFORE CODE: TrustedClient: this.clientToken,
},
});
return data;
}
inference_login() {
return this.request("/auth/trusted", { method:... | Can we check for the http status here? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { withoutRole } from "src/lib/auth";
import { createApiClientFromUser } from "src/lib/oasst_client_factory";
import { getBackendUserCore } from "src/lib/users";
const handler = withoutRole("banned", async (req, res, token) => {
const user = await getBackendUserCore(token.sub);
const oasstApiClient = createA... | You can deconstruct the result here | import { withoutRole } from "src/lib/auth";
import { createApiClientFromUser } from "src/lib/oasst_client_factory";
import { getBackendUserCore } from "src/lib/users";
const handler = withoutRole("banned", async (req, res, token) => {
const user = await getBackendUserCore(token.sub);
const oasstApiClient = createA... | @@ -6,7 +6,7 @@ const handler = withoutRole("banned", async (req, res, token) => {
const user = await getBackendUserCore(token.sub);
const oasstApiClient = createApiClientFromUser(user);
if (req.method === "GET") {
- const tos_acceptance_date = await oasstApiClient.fetch_tos_acceptance(user);
+ const tos... | website/src/pages/api/tos.ts | 9 | TypeScript | 0.071 | suggestion | 35 | 19 | 19 | false | Use trusted clients for inference auth | 2,278 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/tos.ts
PR Title: Use trusted clients for inference auth
BEFORE CODE: import { withoutRole } from "src/lib/auth";
import { createApiClientFromUser } from "src/lib/oasst_client_factory";
import { getBackendUserCore } from "src/lib/users... | You can deconstruct the result here | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
@app.middleware("http")
async def log_exceptions(request: fastapi.Request, call_next):
try:
response = await call_next(request)
except Exception:
logger.exception("Exception in request")
raise
return response
def terminate_server(signum, frame):
logger.warning(f"Signal {signum... | Just a preference really but I feel a one-liner is clearer here
`eos_token = tokenizer.eos_token if hasattr(tokenizer, "eos_token") else ""` |
@app.middleware("http")
async def log_exceptions(request: fastapi.Request, call_next):
try:
response = await call_next(request)
except Exception:
logger.exception("Exception in request")
raise
return response
model_loaded: bool = False
fully_loaded: bool = False
model_input_queue... | @@ -42,6 +47,51 @@ def terminate_server(signum, frame):
model: transformers.PreTrainedModel
tokenizer: transformers.PreTrainedTokenizer
+fully_loaded: bool = False
+model_input_queue: Queue = Queue()
+
+
+def model_thread():
+ request: interface.GenerateStreamRequest
+ output_queue: Queue
+ eos_token = "" | inference/worker/basic_hf_server.py | 26 | Python | 0.571 | suggestion | 143 | 51 | 51 | false | changed basic hf server to support quantization and streaming | 2,293 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | yk | Review this pull request change.
Language: Python
File: inference/worker/basic_hf_server.py
PR Title: changed basic hf server to support quantization and streaming
BEFORE CODE:
@app.middleware("http")
async def log_exceptions(request: fastapi.Request, call_next):
try:
response = await call_next(request... | Just a preference really but I feel a one-liner is clearer here
`eos_token = tokenizer.eos_token if hasattr(tokenizer, "eos_token") else ""` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
# get the world size in deeepspeed
if conf["deepspeed"]:
conf["world_size"] = int(os.getenv("WORLD_SIZE", default="1"))
else:
conf["world_size"] = 1
# Override config from command-line
parser = argparse.ArgumentParser()
for key, value in conf.items():
type_ = type(value... | line is duplicate of previous line
```suggestion
``` |
# get the world size in deeepspeed
if conf["deepspeed"]:
conf["world_size"] = int(os.getenv("WORLD_SIZE", default="1"))
else:
conf["world_size"] = 1
# Override config from command-line
parser = argparse.ArgumentParser()
for key, value in conf.items():
type_ = type(value... | @@ -184,6 +184,8 @@ def main():
init_rng(training_conf)
+ init_rng(training_conf)
+ | model/model_training/trainer_rm.py | 26 | Python | 0.571 | refactor | 54 | 51 | 51 | false | changed basic hf server to support quantization and streaming | 2,293 | LAION-AI/Open-Assistant | 29,137 | Python | dvruette | yk | Review this pull request change.
Language: Python
File: model/model_training/trainer_rm.py
PR Title: changed basic hf server to support quantization and streaming
BEFORE CODE:
# get the world size in deeepspeed
if conf["deepspeed"]:
conf["world_size"] = int(os.getenv("WORLD_SIZE", default="1"))
... | line is duplicate of previous line
```suggestion
``` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import inference
from settings import settings
def terminate_worker(signum, frame):
logger.warning(f"Signal {signum}. Terminating worker...")
sys.exit(0)
def main():
signal.signal(sign... | can be removed with latest HF version (4.28.0.dev0) | import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import inference
from settings import settings
def terminate_worker(signum, frame):
logger.warning(f"Signal {signum}. Terminating worker...")
sys.exit(0)
def main():
signal.signal(sign... | @@ -31,10 +31,11 @@ def main():
if model_config.is_lorem:
tokenizer = None
- elif model_config.is_llama:
- tokenizer: transformers.PreTrainedTokenizer = transformers.LlamaTokenizer.from_pretrained(model_config.model_id)
+ # elif model_config.is_llama:
+ # tokenizer: transformers.PreT... | inference/worker/__main__.py | 26 | Python | 0.286 | suggestion | 51 | 51 | 51 | false | changed basic hf server to support quantization and streaming | 2,293 | LAION-AI/Open-Assistant | 29,137 | Python | dvruette | yk | Review this pull request change.
Language: Python
File: inference/worker/__main__.py
PR Title: changed basic hf server to support quantization and streaming
BEFORE CODE: import websocket
import work
from loguru import logger
from oasst_shared import model_configs
from oasst_shared.schemas import inference
from sett... | can be removed with latest HF version (4.28.0.dev0) | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
name = list(data_config.keys())[0]
# first copy the dict, then remove the size and fraction
kwargs = copy.deepcopy(data_config[name])
kwargs.pop("fraction", None)
kwargs.pop("size", None)
return name, kwargs
else:
return data_config, {}
def get_dataset(
... | Is this special return handling really necessary (`return_datasets_as_list` etc)? It is possible to access the list of datasets via the `datasets` member of a `ConcatDataset` instance. | name = list(data_config.keys())[0]
# first copy the dict, then remove the size and fraction
kwargs = copy.deepcopy(data_config[name])
kwargs.pop("fraction", None)
kwargs.pop("size", None)
return name, kwargs
else:
return data_config, {}
def get_dataset(
... | @@ -365,6 +367,8 @@ def get_dataset(conf, mode="sft"):
if val is not None:
evals[dataset_name] = Subset(val, list(range(min(len(val), conf.eval_size)))) if conf.eval_size else val
+ if return_datasets_as_list: | model/model_training/utils.py | 26 | Python | 0.5 | question | 184 | 51 | 51 | false | add dataset counts script | 2,294 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/utils.py
PR Title: add dataset counts script
BEFORE CODE: name = list(data_config.keys())[0]
# first copy the dict, then remove the size and fraction
kwargs = copy.deepcopy(data_config[name])
kwargs.pop("f... | Is this special return handling really necessary (`return_datasets_as_list` etc)? It is possible to access the list of datasets via the `datasets` member of a `ConcatDataset` instance. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
"""
def preprocess_dialogue(self, text):
"""
trim prefix text to last two pairs
Outlier example Assistant answered empty string:
Assistant: Human: That makes sense, I agree with that, though there are many situations that
aren't considered justice, like sending ... | This are changes in the reward/instructor folder. They are not used for the trainer_rm code. I suggest to remove these changes. | """
def preprocess_dialogue(self, text):
"""
trim prefix text to last two pairs
Outlier example Assistant answered empty string:
Assistant: Human: That makes sense, I agree with that, though there are many situations that
aren't considered justice, like sending ... | @@ -261,11 +261,16 @@ def preprocess_dialogue(self, text):
return "Human:".join(last_two_convo)
return "Human: " + "Human:".join(last_two_convo)
- def __init__(self, split="train", sep_token="<sep>") -> None:
+ def __init__(self, split="train", sep_token="<sep>", mode=None) -> None:
... | model/reward/instructor/rank_datasets.py | 26 | Python | 0.429 | suggestion | 127 | 51 | 51 | false | two more datasets | 2,301 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | mikegarts | Review this pull request change.
Language: Python
File: model/reward/instructor/rank_datasets.py
PR Title: two more datasets
BEFORE CODE: """
def preprocess_dialogue(self, text):
"""
trim prefix text to last two pairs
Outlier example Assistant answered empty string:
Ass... | This are changes in the reward/instructor folder. They are not used for the trainer_rm code. I suggest to remove these changes. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { AnimatedCircles } from "./AnimatedCircles";
import { Container } from "./Container";
export function Hero() {
const { t } = useTranslation(["index", "common"]);
const { colorMode } = useColorMode();
const pTextColor = colorMode === "light" ? "text-gray-600" : "text-white";
const fancyTextGradientClass... | do we also need to have the `pTextColor` as part of the class name just like above? does it look good in dark / light mode? | import { AnimatedCircles } from "./AnimatedCircles";
import { Container } from "./Container";
export function Hero() {
const { t } = useTranslation(["index", "common"]);
const { colorMode } = useColorMode();
const pTextColor = colorMode === "light" ? "text-gray-600" : "text-white";
const fancyTextGradientClass... | @@ -27,6 +30,20 @@ export function Hero() {
</Text>
<Text className={`mt-6 text-lg ${pTextColor}`}>{t("blurb")}</Text>
<Text className={`mt-6 text-lg ${pTextColor}`}>{t("blurb1")}</Text>
+ <Box className="mt-6 flex gap-6"> | website/src/components/Hero.tsx | 26 | TypeScript | 0.5 | question | 123 | 51 | 51 | false | Added CTA buttons to the frontpage | 2,302 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | yk | Review this pull request change.
Language: TypeScript
File: website/src/components/Hero.tsx
PR Title: Added CTA buttons to the frontpage
BEFORE CODE: import { AnimatedCircles } from "./AnimatedCircles";
import { Container } from "./Container";
export function Hero() {
const { t } = useTranslation(["index", "comm... | do we also need to have the `pTextColor` as part of the class name just like above? does it look good in dark / light mode? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
If you find the data useful, please cite the paper.
The data format is very simple -- each line of the jsonl files contains a pair of texts,
one "chosen" and one "rejected".
valid train size : 160780
"""
def preprocess_dialogue(self, text):
"""
trim prefix text to l... | There are now duplicate changes to my 5 minutes earlier submitted PR #2306. | If you find the data useful, please cite the paper.
The data format is very simple -- each line of the jsonl files contains a pair of texts,
one "chosen" and one "rejected".
valid train size : 160780
"""
def preprocess_dialogue(self, text):
"""
trim prefix text to l... | @@ -261,16 +261,11 @@ def preprocess_dialogue(self, text):
return "Human:".join(last_two_convo)
return "Human: " + "Human:".join(last_two_convo)
- def __init__(self, split="train", sep_token="<sep>", mode=None) -> None:
+ def __init__(self, split="train", sep_token="<sep>") -> None:
... | model/reward/instructor/rank_datasets.py | 26 | Python | 0.286 | refactor | 75 | 51 | 51 | false | revert instructor code and fix a bug in anthropic ds parsing | 2,307 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | mikegarts | Review this pull request change.
Language: Python
File: model/reward/instructor/rank_datasets.py
PR Title: revert instructor code and fix a bug in anthropic ds parsing
BEFORE CODE: If you find the data useful, please cite the paper.
The data format is very simple -- each line of the jsonl files cont... | There are now duplicate changes to my 5 minutes earlier submitted PR #2306. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
from model_training.custom_datasets.rank_datasets import AnthropicRLHF, HFDataset
def load_anthroopic_rlhf():
train = AnthropicRLHF(split="train")
validation = AnthropicRLHF(split="test")
print(f"AnthropicRLHF extra dataset size: {len(train)}, {len(validation)}")
return train, validation
def load_... | typo:
```suggestion
def load_anthropic_rlhf():
``` | from model_training.custom_datasets.rank_datasets import AnthropicRLHF, HFDataset
def load_anthropic_rlhf():
train = AnthropicRLHF(split="train")
validation = AnthropicRLHF(split="test")
print(f"AnthropicRLHF extra dataset size: {len(train)}, {len(validation)}")
return train, validation
def load_o... | @@ -1,10 +1,9 @@
-from model_training.custom_datasets.rank_datasets import HFDataset
-from reward.instructor.rank_datasets import AnthropicRLHF
+from model_training.custom_datasets.rank_datasets import AnthropicRLHF, HFDataset
def load_anthroopic_rlhf(): | model/model_training/custom_datasets/extra_rm_datasets.py | 4 | Python | 0.571 | suggestion | 53 | 26 | 26 | false | revert instructor code and fix a bug in anthropic ds parsing | 2,307 | LAION-AI/Open-Assistant | 29,137 | Python | dvruette | mikegarts | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/extra_rm_datasets.py
PR Title: revert instructor code and fix a bug in anthropic ds parsing
BEFORE CODE: from model_training.custom_datasets.rank_datasets import AnthropicRLHF, HFDataset
def load_anthroopic_rlhf():
tr... | typo:
```suggestion
def load_anthropic_rlhf():
``` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
parser.add_argument("--no-deepspeed", dest="deepspeed", action="store_false")
parser.add_argument("--wandb-entity", type=str, default="open-assistant")
parser.add_argument("--rng_seed", type=int, help="rng seed")
parser.set_defaults(deepspeed=False)
if notebook:
args, remaining = parser.par... | I don't think this logic is correct, but I can correct it after merge in the improve_small_rm branch.. | parser.add_argument("--wandb-entity", type=str, default="open-assistant")
parser.add_argument("--rng_seed", type=int, help="rng seed")
parser.set_defaults(deepspeed=False)
if notebook:
args, remaining = parser.parse_known_args(notebook_args)
else:
args, remaining = parser.parse_know... | @@ -158,7 +160,7 @@ def argument_parsing(notebook=False, notebook_args=None):
conf["wandb_entity"] = args.wandb_entity
conf["local_rank"] = args.local_rank
conf["deepspeed"] = args.deepspeed
- if args.rng_seed is not None:
+ if "rng_seed" not in conf and args.rng_seed is not None: | model/model_training/trainer_rm.py | 26 | Python | 0.357 | suggestion | 102 | 51 | 51 | false | fix deepspeed issue on trainer_rm.py, add crossentropy support | 2,321 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | theblackcat102 | Review this pull request change.
Language: Python
File: model/model_training/trainer_rm.py
PR Title: fix deepspeed issue on trainer_rm.py, add crossentropy support
BEFORE CODE: parser.add_argument("--no-deepspeed", dest="deepspeed", action="store_false")
parser.add_argument("--wandb-entity", type=str, defau... | I don't think this logic is correct, but I can correct it after merge in the improve_small_rm branch.. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
def get_metrics(conf, tokenizer):
# the reason behind using a list is that we might want to extend the list of our
# metrics in the future for more thorough evaluation
metrics, preprocess_fns = [evaluate.load("accuracy")], [default_preprocess]
# if any(dataset in QA_DATASETS for dataset in conf.dataset... | hm. why was the dtype specification here removed .. ? | def get_metrics(conf, tokenizer):
# the reason behind using a list is that we might want to extend the list of our
# metrics in the future for more thorough evaluation
metrics, preprocess_fns = [evaluate.load("accuracy")], [default_preprocess]
# if any(dataset in QA_DATASETS for dataset in conf.dataset... | @@ -296,12 +299,14 @@ def get_model(conf, tokenizer, pad_vocab_size_to_multiple_of=16):
if conf.is_reward_model:
if "pythia" in conf.model_name:
- model = GPTNeoXRewardModel.from_pretrained(conf.model_name, cache_dir=conf.cache_dir, torch_dtype=dtype) | model/model_training/utils.py | 26 | Python | 0.429 | question | 53 | 51 | 51 | false | fix deepspeed issue on trainer_rm.py, add crossentropy support | 2,321 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | theblackcat102 | Review this pull request change.
Language: Python
File: model/model_training/utils.py
PR Title: fix deepspeed issue on trainer_rm.py, add crossentropy support
BEFORE CODE: def get_metrics(conf, tokenizer):
# the reason behind using a list is that we might want to extend the list of our
# metrics in the futu... | hm. why was the dtype specification here removed .. ? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import fastapi
import pydantic
import sqlalchemy.orm
import sqlmodel
from loguru import logger
from oasst_inference_server import database, models
from oasst_shared.schemas import inference
class UserChatRepository(pydantic.BaseModel):
session: database.AsyncSession
user_id: str = pydantic.Field(..., min_leng... | could we leave this sorted by date? this has the advantage of being functional regardless of which id type we use. Also perhaps slightly faster but not that much. | import fastapi
import pydantic
import sqlalchemy.orm
import sqlmodel
from loguru import logger
from oasst_inference_server import database, models
from oasst_shared.schemas import inference
class UserChatRepository(pydantic.BaseModel):
session: database.AsyncSession
user_id: str = pydantic.Field(..., min_leng... | @@ -17,7 +17,7 @@ class Config:
async def get_chats(self) -> list[models.DbChat]:
query = sqlmodel.select(models.DbChat)
query = query.where(models.DbChat.user_id == self.user_id)
- query = query.order_by(models.DbChat.created_at.desc())
+ query = query.order_by(models.DbChat.id.des... | inference/server/oasst_inference_server/user_chat_repository.py | 20 | Python | 0.571 | style | 162 | 45 | 45 | false | inference: use uuid v7 for most of table | 2,327 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/user_chat_repository.py
PR Title: inference: use uuid v7 for most of table
BEFORE CODE: import fastapi
import pydantic
import sqlalchemy.orm
import sqlmodel
from loguru import logger
from oasst_inference_server import da... | could we leave this sorted by date? this has the advantage of being functional regardless of which id type we use. Also perhaps slightly faster but not that much. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import datetime
import enum
from uuid import uuid4
import sqlalchemy as sa
import sqlalchemy.dialects.postgresql as pg
from loguru import logger
from oasst_inference_server.settings import settings
from oasst_shared.schemas import inference
from sqlmodel import Field, Relationship, SQLModel
from uuid_extensions import... | is the lambda necessary or could it be `default_factory=uuid7str` ? | import datetime
import enum
from uuid import uuid4
import sqlalchemy as sa
import sqlalchemy.dialects.postgresql as pg
from loguru import logger
from oasst_inference_server.settings import settings
from oasst_shared.schemas import inference
from sqlmodel import Field, Relationship, SQLModel
from uuid_extensions import... | @@ -17,7 +18,7 @@ class WorkerEventType(str, enum.Enum):
class DbWorkerComplianceCheck(SQLModel, table=True):
__tablename__ = "worker_compliance_check"
- id: str = Field(default_factory=lambda: str(uuid4()), primary_key=True)
+ id: str = Field(default_factory=lambda: uuid7str(), primary_key=True) | inference/server/oasst_inference_server/models/worker.py | 21 | Python | 0.429 | question | 67 | 46 | 46 | false | inference: use uuid v7 for most of table | 2,327 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | notmd | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/models/worker.py
PR Title: inference: use uuid v7 for most of table
BEFORE CODE: import datetime
import enum
from uuid import uuid4
import sqlalchemy as sa
import sqlalchemy.dialects.postgresql as pg
from loguru import ... | is the lambda necessary or could it be `default_factory=uuid7str` ? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
from collections import defaultdict
from typing import List
import numpy as np
from datasets import load_dataset
from torch.utils.data import Dataset
SEED = 2020
class SHPDataset(Dataset):
"""
Dataset class to load stanfordnlp/SHP for Reward Modeling
"""
name = "SHP"
def __init__(self, split):... | Propose to just merge rank dataset for webgpt with SFT and RL mode under qa_datasets.py
See [here](https://github.com/LAION-AI/Open-Assistant/blob/rm-fixes/model/model_training/custom_datasets/qa_datasets.py#L194)
| import random
from collections import defaultdict
from typing import List
import numpy as np
from datasets import load_dataset
from torch.utils.data import Dataset
SEED = 2020
class SHPDataset(Dataset):
"""
Dataset class to load stanfordnlp/SHP for Reward Modeling
"""
name = "SHP"
def __init__... | model/model_training/custom_datasets/rank_datasets.py | 1 | Python | 0.429 | suggestion | 218 | 26 | 26 | false | Improve scores of small 1.4B reward model.. | 2,329 | LAION-AI/Open-Assistant | 29,137 | Python | theblackcat102 | andreaskoepf | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/rank_datasets.py
PR Title: Improve scores of small 1.4B reward model..
BEFORE CODE: from collections import defaultdict
from typing import List
import numpy as np
from datasets import load_dataset
from torch.utils.data imp... | Propose to just merge rank dataset for webgpt with SFT and RL mode under qa_datasets.py
See [here](https://github.com/LAION-AI/Open-Assistant/blob/rm-fixes/model/model_training/custom_datasets/qa_datasets.py#L194)
| You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. | |
"""
High level functions for model training
"""
from model_training.custom_datasets.extra_rm_datasets import (
load_anthropic_rlhf,
load_hellaswag,
load_open_ai_summarize_from_feedback,
load_shp,
)
from model_training.custom_datasets.instruction import INSTRUCTION_DATASETS, InstructionDataset
from m... | I would prefer to leave WebGPTQA as WebGPT and remove the WebGPTRank under rank_dataset as the ranking function was supported in the original WebGPT ( under qa_dataset ) | """
High level functions for model training
"""
from typing import Optional
import numpy as np
from model_training.custom_datasets.extra_rm_datasets import load_anthropic_rlhf, load_hellaswag, load_shp
from model_training.custom_datasets.instruction import INSTRUCTION_DATASETS, InstructionDataset
from model_traini... | model/model_training/custom_datasets/__init__.py | 1 | Python | 0.429 | suggestion | 169 | 26 | 26 | false | Improve scores of small 1.4B reward model.. | 2,329 | LAION-AI/Open-Assistant | 29,137 | Python | theblackcat102 | andreaskoepf | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/__init__.py
PR Title: Improve scores of small 1.4B reward model..
BEFORE CODE: """
High level functions for model training
"""
from model_training.custom_datasets.extra_rm_datasets import (
load_anthropic_rlhf,
... | I would prefer to leave WebGPTQA as WebGPT and remove the WebGPTRank under rank_dataset as the ranking function was supported in the original WebGPT ( under qa_dataset ) | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. | |
from model_training.custom_datasets.rank_datasets import AnthropicRLHF, HellaSwagDataset, HFDataset, SHPDataset
from torch.utils.data import Dataset
def load_anthropic_rlhf() -> tuple[Dataset, Dataset]:
train = AnthropicRLHF(split="train")
validation = AnthropicRLHF(split="test")
print(f"AnthropicRLHF dat... | Could we use Tuple instead of tuple? I had some bad backward compatibility on Python3.8 (no other version choice ) with these static typing. | from model_training.custom_datasets.rank_datasets import AnthropicRLHF, HellaSwagDataset, SHPDataset
from torch.utils.data import Dataset
def load_anthropic_rlhf() -> tuple[Dataset, Dataset]:
train = AnthropicRLHF(split="train")
validation = AnthropicRLHF(split="test")
return train, validation
def load_... | model/model_training/custom_datasets/extra_rm_datasets.py | 1 | Python | 0.429 | suggestion | 141 | 26 | 21 | false | Improve scores of small 1.4B reward model.. | 2,329 | LAION-AI/Open-Assistant | 29,137 | Python | theblackcat102 | andreaskoepf | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/extra_rm_datasets.py
PR Title: Improve scores of small 1.4B reward model..
BEFORE CODE: from model_training.custom_datasets.rank_datasets import AnthropicRLHF, HellaSwagDataset, HFDataset, SHPDataset
from torch.utils.data i... | Could we use Tuple instead of tuple? I had some bad backward compatibility on Python3.8 (no other version choice ) with these static typing. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. | |
import { withoutRole } from "src/lib/auth";
import { createApiClientFromUser } from "src/lib/oasst_client_factory";
import { getBackendUserCore } from "src/lib/users";
const handler = withoutRole("banned", async (req, res, token) => {
let userid: string;
if (req.query.uid) {
userid = typeof req.query.uid === "... | Also can you check the current user has any mod or admin role herer? We dont want to expose stats to every one. | import { ROLES } from "src/components/RoleSelect";
import { withoutRole } from "src/lib/auth";
import { createApiClientFromUser } from "src/lib/oasst_client_factory";
import { getBackendUserCore } from "src/lib/users";
const handler = withoutRole("banned", async (req, res, token) => {
let userid: string;
const isM... | @@ -3,7 +3,14 @@ import { createApiClientFromUser } from "src/lib/oasst_client_factory";
import { getBackendUserCore } from "src/lib/users";
const handler = withoutRole("banned", async (req, res, token) => {
- const user = await getBackendUserCore(token.sub);
+ let userid: string;
+ if (req.query.uid) { | website/src/pages/api/user_stats.ts | 7 | TypeScript | 0.429 | suggestion | 111 | 20 | 25 | false | stats are shown on the admin | 2,330 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | vivasvan1 | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/user_stats.ts
PR Title: stats are shown on the admin
BEFORE CODE: import { withoutRole } from "src/lib/auth";
import { createApiClientFromUser } from "src/lib/oasst_client_factory";
import { getBackendUserCore } from "src/lib/users";
... | Also can you check the current user has any mod or admin role herer? We dont want to expose stats to every one. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { boolean } from "boolean";
import { Head, Html, Main, NextScript } from "next/document";
export default function Document() {
return (
<Html className="h-full antialiased" lang="en">
<Head>
<link rel="shortcut icon" type="image/png" href="/images/logos/favicon.png" />
<script
... | is there really no other way?
~~can we move this code into the body of the react function?~~
ok this function is executed server-side only. | import { Head, Html, Main, NextScript } from "next/document";
export default function Document() {
return (
<Html className="h-full antialiased" lang="en">
<Head>
<link rel="shortcut icon" type="image/png" href="/images/logos/favicon.png" />
</Head>
<body className="flex h-full flex-col... | @@ -1,10 +1,27 @@
+import { boolean } from "boolean";
import { Head, Html, Main, NextScript } from "next/document";
export default function Document() {
return (
<Html className="h-full antialiased" lang="en">
<Head>
<link rel="shortcut icon" type="image/png" href="/images/logos/favicon.png"... | website/src/pages/_document.tsx | 10 | TypeScript | 0.429 | refactor | 145 | 35 | 16 | false | Use SSR for every pages | 2,334 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/pages/_document.tsx
PR Title: Use SSR for every pages
BEFORE CODE: import { boolean } from "boolean";
import { Head, Html, Main, NextScript } from "next/document";
export default function Document() {
return (
<Html className="h-full ant... | is there really no other way?
~~can we move this code into the body of the react function?~~
ok this function is executed server-side only. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { boolean } from "boolean";
import { withoutRole } from "src/lib/auth";
import { createInferenceClient } from "src/lib/oasst_inference_client";
import { InferencePostAssistantMessageParams } from "src/types/Chat";
const handler = withoutRole("banned", async (req, res, token) => {
if (!boolean(process.env.ENAB... | this line is repeated too many times, maybe we can re-add the `isChatEnabled` but we can use the new `getEnv()` in it | import { withoutRole } from "src/lib/auth";
import { isChatEnable } from "src/lib/isChatEnable";
import { createInferenceClient } from "src/lib/oasst_inference_client";
import { InferencePostAssistantMessageParams } from "src/types/Chat";
const handler = withoutRole("banned", async (req, res, token) => {
if (!isChat... | @@ -1,8 +1,12 @@
+import { boolean } from "boolean";
import { withoutRole } from "src/lib/auth";
import { createInferenceClient } from "src/lib/oasst_inference_client";
import { InferencePostAssistantMessageParams } from "src/types/Chat";
const handler = withoutRole("banned", async (req, res, token) => {
+ if (!... | website/src/pages/api/chat/assistant_message.ts | 7 | TypeScript | 0.571 | suggestion | 117 | 21 | 21 | false | Use SSR for every pages | 2,334 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/chat/assistant_message.ts
PR Title: Use SSR for every pages
BEFORE CODE: import { boolean } from "boolean";
import { withoutRole } from "src/lib/auth";
import { createInferenceClient } from "src/lib/oasst_inference_client";
import { I... | this line is repeated too many times, maybe we can re-add the `isChatEnabled` but we can use the new `getEnv()` in it | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
self.pairs.append((prefix, human, answer))
else:
self.pairs.append(("", human, answer))
# Does this make sense?
prefix += "{}{}{}{}".format(
"Question:",
... | could be tuple instead of list | if convo_round["round"] > 2:
self.pairs.append((prefix, human, answer))
else:
self.pairs.append(("", human, answer))
# Does this make sense?
prefix += "{}{}{}{}".forma... | @@ -419,37 +420,78 @@ def __getitem__(self, index):
return self.pairs[index]
-class AlpacaBase(Dataset):
- def __init__(self, dataset_name: str, mode: str, cache_dir: str = None) -> None:
+class AlpacaBaseDataset(Dataset):
+ def __init__(self, data: list, mode: str):
super().__init__()
+ ... | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.143 | suggestion | 30 | 51 | 51 | false | Add alpaca reverse augmentation possibility | 2,342 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: Add alpaca reverse augmentation possibility
BEFORE CODE: self.pairs.append((prefix, human, answer))
else:
self.pairs.a... | could be tuple instead of list | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
def __getitem__(self, index):
return self.pairs[index]
class AlpacaBaseDataset(Dataset):
def __init__(self, data: list, mode: str):
super().__init__()
self.data = data
if mode not in ["sft", "rl"]:
raise NotImplementedError(
f"Alpaca Dataset for mode... | If these classes don't add any functionality they could be removed (i.e. base class be used directly). |
def __getitem__(self, index):
return self.pairs[index]
class AlpacaBaseDataset(Dataset):
def __init__(self, data: list, mode: str):
super().__init__()
self.data = data
if mode not in ("sft", "rl"):
raise NotImplementedError(
f"Alpaca Dataset for mod... | @@ -419,37 +420,78 @@ def __getitem__(self, index):
return self.pairs[index]
-class AlpacaBase(Dataset):
- def __init__(self, dataset_name: str, mode: str, cache_dir: str = None) -> None:
+class AlpacaBaseDataset(Dataset):
+ def __init__(self, data: list, mode: str):
super().__init__()
+ ... | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.429 | suggestion | 103 | 51 | 51 | false | Add alpaca reverse augmentation possibility | 2,342 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: Add alpaca reverse augmentation possibility
BEFORE CODE: def __getitem__(self, index):
return self.pairs[index]
class AlpacaBaseDataset(Dataset):
def __init__(self, data: list, mod... | If these classes don't add any functionality they could be removed (i.e. base class be used directly). | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
def __getitem__(self, index):
question, answer = self.data[index]
if self.mode == "sft":
return (question, answer)
elif self.mode == "rl":
return (question,)
class AlpacaDataset(AlpacaBaseDataset):
pass
class CodeAlpacaDataset(AlpacaBaseDataset):
pass
d... | base class type annotation for factory function would be better imo | return len(self.data)
def __getitem__(self, index):
question, answer = self.data[index]
if self.mode == "sft":
return (question, answer)
elif self.mode == "rl":
return (question,)
def load_alpaca_dataset(
dataset_name: str,
val_split: float,
cac... | @@ -419,37 +420,78 @@ def __getitem__(self, index):
return self.pairs[index]
-class AlpacaBase(Dataset):
- def __init__(self, dataset_name: str, mode: str, cache_dir: str = None) -> None:
+class AlpacaBaseDataset(Dataset):
+ def __init__(self, data: list, mode: str):
super().__init__()
+ ... | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.429 | suggestion | 67 | 51 | 51 | false | Add alpaca reverse augmentation possibility | 2,342 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: Add alpaca reverse augmentation possibility
BEFORE CODE:
def __getitem__(self, index):
question, answer = self.data[index]
if self.mode == "sft":
return (question, a... | base class type annotation for factory function would be better imo | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
dataset_name = dataset_name.lower()
if dataset_name in QA_DATASETS:
dataset = QADataset(dataset_name, data_path, "train")
if not dataset.no_val:
eval = QADataset(dataset_name, data_path, "validation")
train = dataset
elif dataset_name in SUMMARIZATION_DATASETS:
... | use immutable tuple instead of list | if mode == "rm":
assert dataset_name in RM_DATASETS, f"Dataset {dataset_name} not supported for reward modeling"
data_path = data_path or conf.cache_dir
dataset_name = dataset_name.lower()
if dataset_name in QA_DATASETS:
dataset = QADataset(dataset_name, data_path, "train")
if ... | @@ -114,10 +113,8 @@ def get_one_dataset(
dataset = DiveMT()
elif dataset_name == "webgpt":
dataset = WebGPT(mode=mode)
- elif dataset_name == "alpaca":
- dataset = Alpaca(mode=mode, cache_dir=data_path)
- elif dataset_name == "code_alpaca":
- dataset = CodeAlpaca(mode=mode, c... | model/model_training/custom_datasets/__init__.py | 26 | Python | 0.143 | suggestion | 35 | 51 | 51 | false | Add alpaca reverse augmentation possibility | 2,342 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/__init__.py
PR Title: Add alpaca reverse augmentation possibility
BEFORE CODE: dataset_name = dataset_name.lower()
if dataset_name in QA_DATASETS:
dataset = QADataset(dataset_name, data_path, "train")
... | use immutable tuple instead of list | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
pairs = self._partition_list(truncated_dialogue, 2)
# add previous history to conversation pairs
# e.g. [[Q1, A1], [Q2, A2]] -> [[Q1, A1], [Q1, A1, Q2, A2]]
pairs = list(accumulate(pairs, lambda x, y: x + y))
return pairs
def __init__(self, cache_dir: str... | TODO: Change return depending on mode | """
Open / close book QA datasets
"""
import copy
import glob
import json
import os
import re
from collections import defaultdict
from pathlib import Path
from urllib.request import urlopen
import numpy as np
from datasets import load_dataset
from torch.utils.data import Dataset
# @agoryuno contributed this
re_re... | @@ -453,3 +456,64 @@ def __init__(self, mode: str = "sft", cache_dir: str = None) -> None:
class CodeAlpaca(AlpacaBase):
def __init__(self, mode: str = "sft", cache_dir: str = None) -> None:
super().__init__(dataset_name="sahil2801/CodeAlpaca-20k", mode=mode, cache_dir=cache_dir)
+
+
+class Vicuna(Datase... | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.143 | suggestion | 37 | 27 | 512 | false | Add vicuna dataset | 2,364 | LAION-AI/Open-Assistant | 29,137 | Python | CloseChoice | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: Add vicuna dataset
BEFORE CODE: pairs = self._partition_list(truncated_dialogue, 2)
# add previous history to conversation pairs
# e.g. [[Q1, A1], [Q2, A2]] -> [[... | TODO: Change return depending on mode | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
Returns:
list[list]: list of sublists with length chunk_size
"""
return [l[i : i + chunk_size] for i in range(0, len(l), chunk_size)]
def process_vicuna_conversations(self, data: dict[str, Any], input_max_length: int) -> list[list[str]] | None:
# Perform some sanity chec... | This is not yet correct, should just be the truncated dialogue | message = line["value"]
if role != speaker:
if role is not None:
dialogue.append("\n".join(messages))
messages = []
role = speaker
messages.append(message.strip())
if role is not None and len(messages) >... | @@ -453,3 +456,64 @@ def __init__(self, mode: str = "sft", cache_dir: str = None) -> None:
class CodeAlpaca(AlpacaBase):
def __init__(self, mode: str = "sft", cache_dir: str = None) -> None:
super().__init__(dataset_name="sahil2801/CodeAlpaca-20k", mode=mode, cache_dir=cache_dir)
+
+
+class Vicuna(Datase... | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.286 | suggestion | 62 | 49 | 41 | false | Add vicuna dataset | 2,364 | LAION-AI/Open-Assistant | 29,137 | Python | CloseChoice | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: Add vicuna dataset
BEFORE CODE: Returns:
list[list]: list of sublists with length chunk_size
"""
return [l[i : i + chunk_size] for i in range(0, len(l), chunk_siz... | This is not yet correct, should just be the truncated dialogue | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
from model_training.custom_datasets.toxic_conversation import ProsocialDialogue, ProsocialDialogueExplaination
from model_training.custom_datasets.translation import WMT2019, DiveMT, TEDTalk
from sklearn.model_selection import train_test_split
from torch.utils.data import Dataset, Subset
QA_DATASETS = list(QADataset.D... | please use ',' at the end to have one entry per line. | from model_training.custom_datasets.toxic_conversation import ProsocialDialogue, ProsocialDialogueExplaination
from model_training.custom_datasets.translation import WMT2019, DiveMT, TEDTalk
from sklearn.model_selection import train_test_split
from torch.utils.data import Dataset, Subset
QA_DATASETS = list(QADataset.D... | @@ -45,13 +46,7 @@
"oa_translated",
]
-RL_DATASETS = [
- "webgpt",
- "private_tuning",
- "alpaca",
- "hf_summary",
- "hf_summary_pairs",
-]
+RL_DATASETS = ["webgpt", "private_tuning", "alpaca", "hf_summary", "hf_summary_pairs", "vicuna"] | model/model_training/custom_datasets/__init__.py | 26 | Python | 0.286 | suggestion | 53 | 51 | 51 | false | Add vicuna dataset | 2,364 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/__init__.py
PR Title: Add vicuna dataset
BEFORE CODE: from model_training.custom_datasets.toxic_conversation import ProsocialDialogue, ProsocialDialogueExplaination
from model_training.custom_datasets.translation import WMT... | please use ',' at the end to have one entry per line. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { useChatContext } from "./ChatContext";
import { ChatConversation } from "./ChatConversation";
const CHAT_CONFIG_KEY = "CHAT_CONFIG";
export const ChatSection = ({ chatId }: { chatId: string }) => {
const { modelInfos } = useChatContext();
const oldConfig = localStorage.getItem(CHAT_CONFIG_KEY);
consol... | can use extract this to an empty component? watching on root component causes a lot rerender. You result the form context, so no need to pass any props. | import { useChatContext } from "./ChatContext";
import { ChatConversation } from "./ChatConversation";
export const ChatSection = ({ chatId }: { chatId: string }) => {
const { modelInfos } = useChatContext();
console.assert(modelInfos.length > 0, "No model config was found");
const form = useForm<ChatConfigForm... | @@ -1,21 +1,35 @@
+import { useEffect } from "react";
import { FormProvider, useForm } from "react-hook-form";
import { ChatConfigForm } from "src/types/Chat";
import { ChatConfigSummary } from "./ChatConfigSummary";
import { useChatContext } from "./ChatContext";
import { ChatConversation } from "./ChatConversa... | website/src/components/Chat/ChatSection.tsx | 26 | TypeScript | 0.429 | refactor | 152 | 35 | 26 | false | Persist chat config | 2,367 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatSection.tsx
PR Title: Persist chat config
BEFORE CODE: import { useChatContext } from "./ChatContext";
import { ChatConversation } from "./ChatConversation";
const CHAT_CONFIG_KEY = "CHAT_CONFIG";
export const ChatSection ... | can use extract this to an empty component? watching on root component causes a lot rerender. You result the form context, so no need to pass any props. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { useEffect } from "react";
import { FormProvider, useForm } from "react-hook-form";
import { ChatConfigForm } from "src/types/Chat";
import { ChatConfigSummary } from "./ChatConfigSummary";
import { useChatContext } from "./ChatContext";
import { ChatConversation } from "./ChatConversation";
const CHAT_CONFIG... | Isn't this crash on the server side? | import { FormProvider, useForm } from "react-hook-form";
import { getCachedChatForm, useCacheChatForm } from "src/hooks/chat/useCacheConfig";
import { ChatConfigForm } from "src/types/Chat";
import { ChatConfigSummary } from "./ChatConfigSummary";
import { useChatContext } from "./ChatContext";
import { ChatConversati... | @@ -1,21 +1,35 @@
+import { useEffect } from "react";
import { FormProvider, useForm } from "react-hook-form";
import { ChatConfigForm } from "src/types/Chat";
import { ChatConfigSummary } from "./ChatConfigSummary";
import { useChatContext } from "./ChatContext";
import { ChatConversation } from "./ChatConversa... | website/src/components/Chat/ChatSection.tsx | 14 | TypeScript | 0.071 | bug | 36 | 39 | 31 | false | Persist chat config | 2,367 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatSection.tsx
PR Title: Persist chat config
BEFORE CODE: import { useEffect } from "react";
import { FormProvider, useForm } from "react-hook-form";
import { ChatConfigForm } from "src/types/Chat";
import { ChatConfigSummary ... | Isn't this crash on the server side? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { getValidDisplayName } from "src/lib/display_name_validation";
import { userlessApiClient } from "src/lib/oasst_client_factory";
import prisma from "src/lib/prismadb";
import { getFrontendUserIdForDiscordUser } from "src/lib/users";
import { LeaderboardEntity, LeaderboardTimeFrame } from "src/types/Leaderboard"... | fix a bug we already have now:
if you go to the prod deployment / admin / open a user that is logged in with discord, the stats are missing at the bottom. | import { getValidDisplayName } from "src/lib/display_name_validation";
import { userlessApiClient } from "src/lib/oasst_client_factory";
import prisma from "src/lib/prismadb";
import { getFrontendUserIdForUser } from "src/lib/users";
import { LeaderboardEntity, LeaderboardTimeFrame } from "src/types/Leaderboard";
impor... | @@ -51,7 +52,8 @@ const ManageUser = ({ user }: InferGetServerSidePropsType<typeof getServerSidePr
const toast = useToast();
const { data: stats } = uswSWRImmutable<Partial<{ [time in LeaderboardTimeFrame]: LeaderboardEntity }>>(
- "/api/user_stats?uid=" + user.id,
+ "/api/user_stats?" +
+ new URLSea... | website/src/pages/admin/manage_user/[id].tsx | 26 | TypeScript | 0.571 | bug | 158 | 51 | 51 | false | Google auth website | 2,391 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/pages/admin/manage_user/[id].tsx
PR Title: Google auth website
BEFORE CODE: import { getValidDisplayName } from "src/lib/display_name_validation";
import { userlessApiClient } from "src/lib/oasst_client_factory";
import prisma from "src/lib/pri... | fix a bug we already have now:
if you go to the prod deployment / admin / open a user that is logged in with discord, the stats are missing at the bottom. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
try:
response = await call_next(request)
except Exception:
logger.exception("Exception in request")
raise
return response
pipeline_loaded: bool = False
pipeline: Blade2Blade
@app.on_event("startup")
async def load_pipeline():
global pipeline_loaded, pipeline
pipeline = Bl... | can we make this into a `while` with `asyncio.sleep`? or perhaps the question would be, how long does it take to init? 1 second? 1 minute? | pipeline_loaded: bool = False
pipeline: Blade2Blade
executor = ThreadPoolExecutor()
@app.on_event("startup")
async def load_pipeline():
global pipeline_loaded, pipeline
pipeline = Blade2Blade(settings.safety_model_name)
# warmup
input = "|prompter|Hey,how are you?|endoftext|"
_ = pipeline.predict(... | @@ -0,0 +1,65 @@
+# A FastAPI server to run the safety pipeline
+
+import fastapi
+import uvicorn
+from blade2blade import Blade2Blade
+from fastapi.middleware.cors import CORSMiddleware
+from loguru import logger
+from oasst_shared.schemas import inference
+from settings import settings
+
+app = fastapi.FastAPI()
+
+
... | inference/safety/main.py | 26 | Python | 0.5 | question | 138 | 41 | 39 | false | Add safety server to inference | 2,449 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | olliestanley | Review this pull request change.
Language: Python
File: inference/safety/main.py
PR Title: Add safety server to inference
BEFORE CODE: try:
response = await call_next(request)
except Exception:
logger.exception("Exception in request")
raise
return response
pipeline_loaded: bool... | can we make this into a `while` with `asyncio.sleep`? or perhaps the question would be, how long does it take to init? 1 second? 1 minute? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
class InternalFinishedMessageResponse(WorkerResponseBase):
response_type: Literal["internal_finished_message"] = "internal_finished_message"
message: MessageRead
class InternalErrorResponse(WorkerResponseBase):
response_type: Literal["internal_error"] = "internal_error"
error: str
message: Messag... | what does `...` mean in this context, is this the same as `pass`?
I am not a python guru so bare with me on this one |
class PongResponse(WorkerResponseBase):
response_type: Literal["pong"] = "pong"
metrics: WorkerMetricsInfo | None = None
class TokenResponse(WorkerResponseBase):
response_type: Literal["token"] = "token"
text: str
log_prob: float | None
token_id: int
class GeneratedTextResponse(WorkerRespo... | @@ -256,6 +256,19 @@ class GeneralErrorResponse(WorkerResponseBase):
error: str
+class SafetyParameters(pydantic.BaseModel):
+ ... | oasst-shared/oasst_shared/schemas/inference.py | 26 | Python | 0.5 | question | 119 | 51 | 51 | false | Add safety server to inference | 2,449 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | olliestanley | Review this pull request change.
Language: Python
File: oasst-shared/oasst_shared/schemas/inference.py
PR Title: Add safety server to inference
BEFORE CODE:
class InternalFinishedMessageResponse(WorkerResponseBase):
response_type: Literal["internal_finished_message"] = "internal_finished_message"
message: ... | what does `...` mean in this context, is this the same as `pass`?
I am not a python guru so bare with me on this one | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
# A FastAPI server to run the safety pipeline
import fastapi
import uvicorn
from blade2blade import Blade2Blade
from fastapi.middleware.cors import CORSMiddleware
from loguru import logger
from oasst_shared.schemas import inference
from settings import settings
app = fastapi.FastAPI()
# Allow CORS
app.add_middlewar... | this is probably not needed, since no browser will try to reach this directly, right? | # A FastAPI server to run the safety pipeline
import asyncio
from concurrent.futures import ThreadPoolExecutor
import fastapi
import uvicorn
from blade2blade import Blade2Blade
from loguru import logger
from oasst_shared.schemas import inference
from settings import settings
app = fastapi.FastAPI()
@app.middleware... | @@ -0,0 +1,65 @@
+# A FastAPI server to run the safety pipeline
+
+import fastapi
+import uvicorn
+from blade2blade import Blade2Blade
+from fastapi.middleware.cors import CORSMiddleware
+from loguru import logger
+from oasst_shared.schemas import inference
+from settings import settings
+
+app = fastapi.FastAPI()
+
+
... | inference/safety/main.py | 21 | Python | 0.214 | question | 85 | 46 | 46 | false | Add safety server to inference | 2,449 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | olliestanley | Review this pull request change.
Language: Python
File: inference/safety/main.py
PR Title: Add safety server to inference
BEFORE CODE: # A FastAPI server to run the safety pipeline
import fastapi
import uvicorn
from blade2blade import Blade2Blade
from fastapi.middleware.cors import CORSMiddleware
from loguru impor... | this is probably not needed, since no browser will try to reach this directly, right? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
logger.exception("Exception in request")
raise
return response
pipeline_loaded: bool = False
pipeline: Blade2Blade
@app.on_event("startup")
async def load_pipeline():
global pipeline_loaded, pipeline
pipeline = Blade2Blade(settings.safety_model_name)
# warmup
input = "|prompter|H... | how does the parallelism of this look like? does the async make it absolutely single-threaded/event-looped? | except Exception:
logger.exception("Exception in request")
raise
return response
pipeline_loaded: bool = False
pipeline: Blade2Blade
executor = ThreadPoolExecutor()
@app.on_event("startup")
async def load_pipeline():
global pipeline_loaded, pipeline
pipeline = Blade2Blade(settings.sa... | @@ -0,0 +1,57 @@
+# A FastAPI server to run the safety pipeline
+
+import asyncio
+
+import fastapi
+import uvicorn
+from blade2blade import Blade2Blade
+from loguru import logger
+from oasst_shared.schemas import inference
+from settings import settings
+
+app = fastapi.FastAPI()
+
+
+@app.middleware("http")
+async de... | inference/safety/main.py | 26 | Python | 0.5 | question | 107 | 39 | 45 | false | Add safety server to inference | 2,449 | LAION-AI/Open-Assistant | 29,137 | Python | yk | olliestanley | Review this pull request change.
Language: Python
File: inference/safety/main.py
PR Title: Add safety server to inference
BEFORE CODE: logger.exception("Exception in request")
raise
return response
pipeline_loaded: bool = False
pipeline: Blade2Blade
@app.on_event("startup")
async def load_pi... | how does the parallelism of this look like? does the async make it absolutely single-threaded/event-looped? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
# construct prompt
messages = [_prepare_message(message) for message in work_request.thread.messages]
prompt = "".join(messages) + V2_ASST_PREFIX
parameters = interface.GenerateStreamParameters.from_work_parameters(work_request.parameters)
if settings.use_stop_sequences:
parameters.stop = ... | we definitely need to track when this happens and feed it back in the response to the backend, and then to the user | # construct prompt
messages = [_prepare_message(message) for message in work_request.thread.messages]
prompt = "".join(messages) + V2_ASST_PREFIX
parameters = interface.GenerateStreamParameters.from_work_parameters(work_request.parameters)
if settings.use_stop_sequences:
parameters.stop = ... | @@ -79,6 +80,18 @@ def _prepare_message(message: inference.MessageRead) -> str:
return prompt, parameters
+def prepare_safe_prompt(prompt: str, safety_opinion: str):
+ safety_opinion = re.sub(r"<pad>|</s>", "", safety_opinion).split("<sep>")
+ label, rots = safety_opinion[0], "and".join([x.strip(".") for... | inference/worker/work.py | 26 | Python | 0.429 | suggestion | 115 | 51 | 51 | false | Add safety server to inference | 2,449 | LAION-AI/Open-Assistant | 29,137 | Python | yk | olliestanley | Review this pull request change.
Language: Python
File: inference/worker/work.py
PR Title: Add safety server to inference
BEFORE CODE: # construct prompt
messages = [_prepare_message(message) for message in work_request.thread.messages]
prompt = "".join(messages) + V2_ASST_PREFIX
parameters = inte... | we definitely need to track when this happens and feed it back in the response to the backend, and then to the user | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
timeout = "timeout"
class MessageRead(pydantic.BaseModel):
id: str
parent_id: str | None
content: str | None
created_at: datetime
role: Literal["prompter", "assistant"]
state: MessageState
score: int
reports: list[Report] = []
# work parameters will be None on user prompts
... | would set to `False` by default bc if this gets merged & released without the safety server being up yet, we want it to be "backwards compatible" | timeout = "timeout"
class MessageRead(pydantic.BaseModel):
id: str
parent_id: str | None
content: str | None
created_at: datetime
role: Literal["prompter", "assistant"]
state: MessageState
score: int
reports: list[Report] = []
# work parameters will be None on user prompts
... | @@ -178,6 +178,19 @@ class Thread(pydantic.BaseModel):
messages: list[MessageRead]
+class SafetyParameters(pydantic.BaseModel):
+ enabled: bool = True | oasst-shared/oasst_shared/schemas/inference.py | 26 | Python | 0.5 | suggestion | 145 | 51 | 51 | false | Add safety server to inference | 2,449 | LAION-AI/Open-Assistant | 29,137 | Python | yk | olliestanley | Review this pull request change.
Language: Python
File: oasst-shared/oasst_shared/schemas/inference.py
PR Title: Add safety server to inference
BEFORE CODE: timeout = "timeout"
class MessageRead(pydantic.BaseModel):
id: str
parent_id: str | None
content: str | None
created_at: datetime
rol... | would set to `False` by default bc if this gets merged & released without the safety server being up yet, we want it to be "backwards compatible" | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
user.tos_acceptance_date = utcnow()
self.db.add(user)
return user
@managed_tx_method(CommitMode.COMMIT)
def mark_user_deleted(self, id: UUID) -> None:
"""
Update a user by global user ID to set deleted flag. Only trusted clients may delete users.
User deleti... | I see the intention is it to keep this info coherent for inference and backend user data.
like it is used in inference/server/oasst_inference_server/routes/admin.py as well.
Can this be defined in oasst-shared/utils.py. ?
example
` DELETED_USER = "Deleted User" `
`user.display_name = utils.DELETED_USER`
Simil... | if tos_acceptance:
user.tos_acceptance_date = utcnow()
self.db.add(user)
return user
@managed_tx_method(CommitMode.COMMIT)
def mark_user_deleted(self, id: UUID) -> None:
"""
Update a user by global user ID to set deleted flag. Only trusted clients may delete... | @@ -119,6 +120,12 @@ def mark_user_deleted(self, id: UUID) -> None:
user.deleted = True
+ # Anonymise user data
+ user.display_name = "Deleted User" | backend/oasst_backend/user_repository.py | 26 | Python | 0.643 | suggestion | 424 | 51 | 51 | false | Add support for user deletion | 2,486 | LAION-AI/Open-Assistant | 29,137 | Python | melvinebenezer | olliestanley | Review this pull request change.
Language: Python
File: backend/oasst_backend/user_repository.py
PR Title: Add support for user deletion
BEFORE CODE: user.tos_acceptance_date = utcnow()
self.db.add(user)
return user
@managed_tx_method(CommitMode.COMMIT)
def mark_user_deleted(se... | I see the intention is it to keep this info coherent for inference and backend user data.
like it is used in inference/server/oasst_inference_server/routes/admin.py as well.
Can this be defined in oasst-shared/utils.py. ?
example
` DELETED_USER = "Deleted User" `
`user.display_name = utils.DELETED_USER`
Simil... | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
}
setQueueInfo(null);
setResponse(null);
setIsSending.off();
},
[getFormValues, setIsSending]
);
const sendPrompterMessage = useCallback(async () => {
const content = inputRef.current?.value.trim();
if (!content) {
return;
}
setIsSending.on();
// find last ... | To prevent refresh, we simply can check whether the chatIdProps is null or not here. | const parent = messages.find((m) => m.id === parentId);
if (!parent) {
// we should never reach here
return console.error("Parent message not found", parentId);
}
if (parent!.state !== "complete") {
return toast({
title: "You are trying reply to a message that i... | @@ -119,27 +111,28 @@ export const ChatConversation = memo(function ChatConversation({ chatId: chatIdP
.find((m) => m.role === "assistant" && m.state === "complete")?.id ?? null;
const prompter_arg: InferencePostPrompterMessageParams = {
- chat_id: currentChatId,
+ chat_id: chatId,
cont... | website/src/components/Chat/ChatConversation.tsx | 26 | TypeScript | 0.357 | suggestion | 84 | 51 | 51 | false | Don't refresh the page after each message | 2,491 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConversation.tsx
PR Title: Don't refresh the page after each message
BEFORE CODE: }
setQueueInfo(null);
setResponse(null);
setIsSending.off();
},
[getFormValues, setIsSending]
);
const sen... | To prevent refresh, we simply can check whether the chatIdProps is null or not here. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
return prepare_safe_prompt(prompt, label, rots)
elif "intervention" in label and safety_level > 0:
return prepare_safe_prompt(prompt, label, rots)
else:
return prompt
def handle_work_request(
ws: websocket.WebSocket,
tokenizer: transformers.PreTrainedTokenizer,
work_request... | could we get as much as possible of the safety response in here? like what was the detected intervention level? | label, rots = safety_opinion[0], "and".join([x.strip(".") for x in safety_opinion[1:]])
label = label.replace("<pad>", "").strip()
return label, rots
def handle_work_request(
ws: websocket.WebSocket,
tokenizer: transformers.PreTrainedTokenizer,
work_request: inference.WorkRequest,
worker_c... | @@ -115,7 +115,18 @@ def handle_work_request(
if settings.enable_safety and work_request.safety_parameters.level:
safety_request = inference.SafetyRequest(inputs=prompt, parameters=work_request.safety_parameters)
safety_response = get_safety_server_response(safety_request)
- prompt = get_s... | inference/worker/work.py | 26 | Python | 0.429 | question | 111 | 51 | 51 | false | Inference worker inform backend on safety intervention | 2,505 | LAION-AI/Open-Assistant | 29,137 | Python | yk | olliestanley | Review this pull request change.
Language: Python
File: inference/worker/work.py
PR Title: Inference worker inform backend on safety intervention
BEFORE CODE: return prepare_safe_prompt(prompt, label, rots)
elif "intervention" in label and safety_level > 0:
return prepare_safe_prompt(prompt, lab... | could we get as much as possible of the safety response in here? like what was the detected intervention level? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
mutateChats(
(chatResponse) => ({
...chatResponse,
chats: chatResponse?.chats.map((chat) => (chat.id === chatId ? { ...chat, title } : chat)) || [],
}),
false
);
},
[mutateChats]
);
return (
<Card
w="260px"
py="3"
gap="1"
h... | is this necessary? I assume we update the title in the `ChatItem`, and then we just refresh here? do we need to re-assign the title here again?. | {
fallbackData: chats,
}
);
const { t } = useTranslation(["common", "chat"]);
const sideProps: CardProps = useMemo(
() =>
isSideBar
? {
w: "260px",
position: "fixed",
}
: {},
[isSideBar]
);
const handleUpdateTitle = useCallback(
... | @@ -15,11 +15,24 @@ import { HEADER_HEIGHT } from "../Header/Header";
import { ChatListItem } from "./ChatListItem";
export const ChatListBase = memo(function ChatListBase(props: CardProps) {
- const { data: chats } = useSWR<GetChatsResponse>(API_ROUTES.LIST_CHAT, get, {
+ const { data: chats, mutate: mutateChats... | website/src/components/Chat/ChatListBase.tsx | 26 | TypeScript | 0.5 | question | 144 | 51 | 51 | false | implement update chat title UI | 2,507 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatListBase.tsx
PR Title: implement update chat title UI
BEFORE CODE: mutateChats(
(chatResponse) => ({
...chatResponse,
chats: chatResponse?.chats.map((chat) => (chat.id === chatId ? { ...chat... | is this necessary? I assume we update the title in the `ChatItem`, and then we just refresh here? do we need to re-assign the title here again?. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
);
const handleConfirmEdit = useCallback(async () => {
const title = inputRef.current?.value;
if (!title) return;
await updateChatTitle({ title, chat_id: chat.id });
setIsEditing.off();
onUpdateTitle({ chatId: chat.id, title });
}, [chat.id, onUpdateTitle, setIsEditing, updateChatTitle]);
c... | Does this work? keep it as `Link` & remove the href when editing (or set to '#'), then it would not navigate. Assuming that was the problem. | );
const handleConfirmEdit = useCallback(async () => {
const title = inputRef.current?.value.trim();
if (!title) return;
await updateChatTitle({ title, chat_id: chat.id });
setIsEditing.off();
onUpdateTitle({ chatId: chat.id, title });
}, [chat.id, onUpdateTitle, setIsEditing, updateChatTitle]... | @@ -1,41 +1,119 @@
-import { Box, Button, Flex, Tooltip, useToast } from "@chakra-ui/react";
-import { LucideIcon, Pencil, Trash2 } from "lucide-react";
+import { Box, Button, CircularProgress, Flex, Input, Tooltip, useBoolean, useOutsideClick } from "@chakra-ui/react";
+import { Check, LucideIcon, Pencil, Trash2, X } ... | website/src/components/Chat/ChatListItem.tsx | 26 | TypeScript | 0.571 | question | 140 | 51 | 51 | false | implement update chat title UI | 2,507 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatListItem.tsx
PR Title: implement update chat title UI
BEFORE CODE: );
const handleConfirmEdit = useCallback(async () => {
const title = inputRef.current?.value;
if (!title) return;
await updateChatTitle({ tit... | Does this work? keep it as `Link` & remove the href when editing (or set to '#'), then it would not navigate. Assuming that was the problem. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
max_new_tokens: 256,
repetition_penalty: 1.2,
temperature: 1,
top_k: 50,
top_p: 0.95,
typical_p: 0.5,
};
const parameterLabel: Record<keyof SamplingParameters, string> = {
max_new_tokens: "Max new tokens",
top_k: "Top K",
top_p: "Top P",
temperature: "Temperature",
repetition_penalty: "Repetition... | should `presets[0].name` be `customPresetName`? | max_new_tokens: 256,
repetition_penalty: 1.2,
temperature: 1,
top_k: 50,
top_p: 0.95,
typical_p: 0.5,
};
const parameterLabel: Record<keyof SamplingParameters, string> = {
max_new_tokens: "Max new tokens",
top_k: "Top K",
top_p: "Top P",
temperature: "Temperature",
repetition_penalty: "Repetition... | @@ -118,7 +118,9 @@ const ChatConfigForm = () => {
const { control, register, reset, getValues } = useFormContext<ChatConfigFormData>();
const selectedModel = useWatch({ name: "model_config_name", control: control });
const presets = modelInfos.find((model) => model.name === selectedModel)!.parameter_configs;
... | website/src/components/Chat/ChatConfigDrawer.tsx | 26 | TypeScript | 0.286 | question | 47 | 51 | 51 | false | fix custom preset selection | 2,511 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConfigDrawer.tsx
PR Title: fix custom preset selection
BEFORE CODE: max_new_tokens: 256,
repetition_penalty: 1.2,
temperature: 1,
top_k: 50,
top_p: 0.95,
typical_p: 0.5,
};
const parameterLabel: Record<keyof Sam... | should `presets[0].name` be `customPresetName`? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
return fastapi.Response(status_code=200)
except Exception:
logger.exception("Error adding report")
return fastapi.Response(status_code=500)
@router.put("/{chat_id}/title")
async def handle_update_title(
chat_id: str,
request: chat_schema.ChatUpdateTitleRequest,
ucr: deps.UserCh... | can we make this a toggle? i.e. hide / unhide, either as a boolean parameter (default hidden) or simply inverting whatever is the database.
Motivation: users would expect the ability to make stuff visible again, unlike delete, which is gone permanently.
| return fastapi.Response(status_code=200)
except Exception:
logger.exception("Error adding report")
return fastapi.Response(status_code=500)
@router.put("/{chat_id}/title")
async def handle_update_title(
chat_id: str,
request: chat_schema.ChatUpdateTitleRequest,
ucr: deps.UserCh... | @@ -270,3 +271,12 @@ async def handle_update_title(
except Exception:
logger.exception("Error when updating chat title")
return fastapi.Response(status_code=500)
+
+
+@router.put("/{chat_id}/hide")
+async def hide_chat(
+ chat_id: str,
+ ucr: UserChatRepository = Depends(deps.create_user_ch... | inference/server/oasst_inference_server/routes/chats.py | 26 | Python | 0.5 | question | 258 | 28 | 29 | false | Add ability to hide (soft delete) chats to inference backend | 2,512 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | olliestanley | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/routes/chats.py
PR Title: Add ability to hide (soft delete) chats to inference backend
BEFORE CODE: return fastapi.Response(status_code=200)
except Exception:
logger.exception("Error adding report")
... | can we make this a toggle? i.e. hide / unhide, either as a boolean parameter (default hidden) or simply inverting whatever is the database.
Motivation: users would expect the ability to make stuff visible again, unlike delete, which is gone permanently.
| You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
id=self.id,
parent_id=self.parent_id,
chat_id=self.chat_id,
content=self.content,
created_at=self.created_at,
role=self.role,
state=self.state,
score=self.score,
work_parameters=self.work_parameters,
... | quick question: what is the motivation of making it an index?
is there a situation where we can have the same id but once visible once not? or is it just about speeding up the queries? | id=self.id,
parent_id=self.parent_id,
chat_id=self.chat_id,
content=self.content,
created_at=self.created_at,
role=self.role,
state=self.state,
score=self.score,
work_parameters=self.work_parameters,
... | @@ -74,12 +74,15 @@ class DbChat(SQLModel, table=True):
messages: list[DbMessage] = Relationship(back_populates="chat")
+ hidden: bool = Field(False, index=True) | inference/server/oasst_inference_server/models/chat.py | 26 | Python | 0.357 | question | 187 | 51 | 51 | false | Add ability to hide (soft delete) chats to inference backend | 2,512 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | olliestanley | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/models/chat.py
PR Title: Add ability to hide (soft delete) chats to inference backend
BEFORE CODE: id=self.id,
parent_id=self.parent_id,
chat_id=self.chat_id,
content=self.... | quick question: what is the motivation of making it an index?
is there a situation where we can have the same id but once visible once not? or is it just about speeding up the queries? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
label = label.replace("<pad>", "").strip()
if "caution" in label and safety_level > 1:
return prepare_safe_prompt(prompt, label, rots)
elif "intervention" in label and safety_level > 0:
return prepare_safe_prompt(prompt, label, rots)
else:
return prompt
def handle_work_request... | We are already receiving a desired safety level from the backend (which in turn comes from the frontend as a user setting) as part of the `SafetyParameters` in the work request. So we can remove the `safety_level` setting from `settings.py` and this line can become:
`prompt = get_safety_opinion(prompt, safety_respon... | label = label.replace("<pad>", "").strip()
if "caution" in label and safety_level > 1:
return prepare_safe_prompt(prompt, label, rots)
elif "intervention" in label and safety_level > 0:
return prepare_safe_prompt(prompt, label, rots)
else:
return prompt
def handle_work_request... | @@ -107,7 +115,7 @@ def handle_work_request(
if settings.enable_safety and work_request.safety_parameters.level:
safety_request = inference.SafetyRequest(inputs=prompt, parameters=work_request.safety_parameters)
safety_response = get_safety_server_response(safety_request)
- prompt = prepar... | inference/worker/work.py | 26 | Python | 0.643 | suggestion | 370 | 51 | 51 | false | Safety level control | 2,516 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | shahules786 | Review this pull request change.
Language: Python
File: inference/worker/work.py
PR Title: Safety level control
BEFORE CODE: label = label.replace("<pad>", "").strip()
if "caution" in label and safety_level > 1:
return prepare_safe_prompt(prompt, label, rots)
elif "intervention" in label and sa... | We are already receiving a desired safety level from the backend (which in turn comes from the frontend as a user setting) as part of the `SafetyParameters` in the work request. So we can remove the `safety_level` setting from `settings.py` and this line can become:
`prompt = get_safety_opinion(prompt, safety_respon... | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
CV = "cv" # Chuvash
KW = "kw" # Cornish
CO = "co" # Corsican
CR = "cr" # Cree
HR = "hr" # Croatian
CS = "cs" # Czech
DA = "da" # Danish
DV = "dv" # Divehi, Dhivehi, Maldivian
NL = "nl" # Dutch, Flemish
DZ = "dz" # Dzongkha
EN = "en" # English
EO = "eo" # Esper... | Does this mean that RM training will not work after this is merged? | return f"{system_tag}{QA_SPECIAL_TOKENS['Question']}{self.questions[0]}{QA_SPECIAL_TOKENS['Answer']}"
else:
return f"{QA_SPECIAL_TOKENS['Question']}{self.questions[0]}{QA_SPECIAL_TOKENS['Answer']}"
if system_tag is not None:
qa_list = [system_tag]
... | @@ -15,20 +28,291 @@ def format_system_prefix(prefix, eos_token):
)
-def format_pairs(pairs, eos_token, add_initial_reply_token=False):
- conversations = [
- "{}{}{}".format(QA_SPECIAL_TOKENS["Question" if i % 2 == 0 else "Answer"], pairs[i], eos_token)
- for i in range(len(pairs))
- ]
- ... | model/model_training/custom_datasets/formatting.py | 26 | Python | 0.214 | question | 67 | 51 | 51 | false | Feature/dataset entry | 2,520 | LAION-AI/Open-Assistant | 29,137 | Python | dvruette | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/formatting.py
PR Title: Feature/dataset entry
BEFORE CODE: CV = "cv" # Chuvash
KW = "kw" # Cornish
CO = "co" # Corsican
CR = "cr" # Cree
HR = "hr" # Croatian
CS = "cs" # Czech
DA = "da" # ... | Does this mean that RM training will not work after this is merged? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
@unique
class Language(str, Enum):
AB = "ab" # Abkhazian
AA = "aa" # Afar
AF = "af" # Afrikaans
AK = "ak" # Akan
SQ = "sq" # Albanian
AM = "am" # Amharic
AR = "ar" # Arabic
AN = "an" # Aragonese
HY = "hy" # Armenian
AS = "as" # Assamese
AV = "av" # Avaric
AE =... | Should the values be quantized here or are they already within a certain set? (I think real values could be difficult to understand for the model) | length: int | None
quality: float | None
humor: float | None
creativity: float | None
@validator("length")
def above_zero(cls, v) -> int:
if v is not None and v < 0:
raise ValueError(f"Length cannot be lower than 0. Received {v}")
return v
@validator("quality", ... | @@ -15,20 +28,291 @@ def format_system_prefix(prefix, eos_token):
)
-def format_pairs(pairs, eos_token, add_initial_reply_token=False):
- conversations = [
- "{}{}{}".format(QA_SPECIAL_TOKENS["Question" if i % 2 == 0 else "Answer"], pairs[i], eos_token)
- for i in range(len(pairs))
- ]
- ... | model/model_training/custom_datasets/formatting.py | 26 | Python | 0.429 | suggestion | 146 | 51 | 51 | false | Feature/dataset entry | 2,520 | LAION-AI/Open-Assistant | 29,137 | Python | dvruette | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/formatting.py
PR Title: Feature/dataset entry
BEFORE CODE:
@unique
class Language(str, Enum):
AB = "ab" # Abkhazian
AA = "aa" # Afar
AF = "af" # Afrikaans
AK = "ak" # Akan
SQ = "sq" # Albanian
... | Should the values be quantized here or are they already within a certain set? (I think real values could be difficult to understand for the model) | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
QA_SPECIAL_TOKENS = {
"Question": "<|prompter|>",
"Answer": "<|assistant|>",
"System": "<|system|>",
"StartPrefix": "<|prefix_begin|>",
"EndPrefix": "<|prefix_end|>",
}
class Mode(Enum):
sft = "sft"
rm = "rm"
rl = "rl"
def format_system_prefix(prefix, eos_token):
return "{}{}{}"... | worth considering extracting this class into a separate file. not necessary, i'll leave it up to you |
QA_SPECIAL_TOKENS = {
"Question": "<|prompter|>",
"Answer": "<|assistant|>",
"System": "<|system|>",
"StartPrefix": "<|prefix_begin|>",
"EndPrefix": "<|prefix_end|>",
}
def format_system_prefix(prefix, eos_token):
return "{}{}{}".format(
QA_SPECIAL_TOKENS["System"],
prefix,
... | @@ -15,20 +28,291 @@ def format_system_prefix(prefix, eos_token):
)
-def format_pairs(pairs, eos_token, add_initial_reply_token=False):
- conversations = [
- "{}{}{}".format(QA_SPECIAL_TOKENS["Question" if i % 2 == 0 else "Answer"], pairs[i], eos_token)
- for i in range(len(pairs))
- ]
- ... | model/model_training/custom_datasets/formatting.py | 26 | Python | 0.429 | refactor | 100 | 51 | 51 | false | Feature/dataset entry | 2,520 | LAION-AI/Open-Assistant | 29,137 | Python | dvruette | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/formatting.py
PR Title: Feature/dataset entry
BEFORE CODE:
QA_SPECIAL_TOKENS = {
"Question": "<|prompter|>",
"Answer": "<|assistant|>",
"System": "<|system|>",
"StartPrefix": "<|prefix_begin|>",
"EndPre... | worth considering extracting this class into a separate file. not necessary, i'll leave it up to you | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
from loguru import logger
from oasst_inference_server import database, models
from oasst_shared import utils as shared_utils
async def delete_user_from_db(session: database.AsyncSession, user_id: str) -> bool:
"""Deletes a user."""
logger.info(f"Deleting user {user_id}")
user = await session.get(models.Db... | I think we should just raise an exception (404) on fail here instead of returning `bool`, that's more in line with the rest of the codebase | import fastapi
from loguru import logger
from oasst_inference_server import database, models
from oasst_shared import utils as shared_utils
async def delete_user_from_db(session: database.AsyncSession, user_id: str):
"""Deletes a user."""
logger.info(f"Deleting user {user_id}")
user = await session.get(mo... | @@ -0,0 +1,20 @@
+from loguru import logger
+from oasst_inference_server import database, models
+from oasst_shared import utils as shared_utils
+
+
+async def delete_user_from_db(session: database.AsyncSession, user_id: str) -> bool:
+ """Deletes a user."""
+ logger.info(f"Deleting user {user_id}")
+ user = a... | inference/server/oasst_inference_server/admin.py | 11 | Python | 0.714 | bug | 139 | 21 | 24 | false | Add delete account page | 2,523 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | AbdBarho | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/admin.py
PR Title: Add delete account page
BEFORE CODE: from loguru import logger
from oasst_inference_server import database, models
from oasst_shared import utils as shared_utils
async def delete_user_from_db(session... | I think we should just raise an exception (404) on fail here instead of returning `bool`, that's more in line with the rest of the codebase | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
if (req.method !== "DELETE") {
return res.status(400).end();
}
console.log("deleting user", token.sub);
try {
const backendUserCore = await getBackendUserCore(token.sub);
const client = createApiClientFromUser(backendUserCore);
await client.delete_account(backendUserCore);
console.log(`user... | I think we need to setup the logger for frontend soon, since console.log is blocking the cpu | if (req.method !== "DELETE") {
return res.status(400).end();
}
console.log("deleting user", token.sub);
try {
const backendUserCore = await getBackendUserCore(token.sub);
const client = createApiClientFromUser(backendUserCore);
await client.delete_account(backendUserCore);
console.log(`user... | @@ -0,0 +1,55 @@
+import { AxiosError } from "axios";
+import type { NextApiRequest, NextApiResponse } from "next";
+import { getToken } from "next-auth/jwt";
+import { createApiClientFromUser } from "src/lib/oasst_client_factory";
+import { createInferenceClient } from "src/lib/oasst_inference_client";
+import prisma ... | website/src/pages/api/account/delete.ts | 26 | TypeScript | 0.214 | suggestion | 92 | 42 | 42 | false | Add delete account page | 2,523 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/account/delete.ts
PR Title: Add delete account page
BEFORE CODE: if (req.method !== "DELETE") {
return res.status(400).end();
}
console.log("deleting user", token.sub);
try {
const backendUserCore = await getBackendUs... | I think we need to setup the logger for frontend soon, since console.log is blocking the cpu | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
cache_dir=cache_dir,
data_files=["ShareGPT_V3_unfiltered_cleaned_split.json"],
revision="192ab2185289094fc556ec8ce5ce1e8e587154ca",
)["train"]
for data in dataset:
if (
processed_data := self.process_vicuna_conversations(data, input_max_len... | could we please also use the same regex that is defined here: https://github.com/LAION-AI/Open-Assistant/blob/2e6f22c2a137ee60ef68cf908073c1e29babd633/model/model_training/custom_datasets/qa_datasets.py#L20
maybe also add upper case single latter references? | cache_dir=cache_dir,
data_files=["ShareGPT_V3_unfiltered_cleaned_split.json"],
revision="192ab2185289094fc556ec8ce5ce1e8e587154ca",
)["train"]
for data in dataset:
if (
processed_data := self.process_vicuna_conversations(data, input_max_len... | @@ -552,6 +552,7 @@ class DatabricksDolly15k(Dataset):
def __init__(self, cache_dir: str | Path, mode: str = "sft", input_max_length: int = 2048) -> None:
super().__init__()
self.rows = []
+ self.citation_regex = re.compile(r"\[[a-z]\]") # removes citations in the form of e.g. [a] | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.5 | question | 261 | 51 | 51 | false | remove references from dolly | 2,535 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: remove references from dolly
BEFORE CODE: cache_dir=cache_dir,
data_files=["ShareGPT_V3_unfiltered_cleaned_split.json"],
revision="192ab2185289094fc556ec8ce5ce1e8... | could we please also use the same regex that is defined here: https://github.com/LAION-AI/Open-Assistant/blob/2e6f22c2a137ee60ef68cf908073c1e29babd633/model/model_training/custom_datasets/qa_datasets.py#L20
maybe also add upper case single latter references? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
try:
await ucr.update_score(message_id=message_id, score=vote_request.score)
return fastapi.Response(status_code=200)
except Exception:
logger.exception("Error adding vote")
return fastapi.Response(status_code=500)
@router.post("/{chat_id}/messages/{message_id}/reports")
async ... | Maybe "/{chat_id/}" is enough? | try:
await ucr.update_score(message_id=message_id, score=vote_request.score)
return fastapi.Response(status_code=200)
except Exception:
logger.exception("Error adding vote")
return fastapi.Response(status_code=500)
@router.post("/{chat_id}/messages/{message_id}/reports")
async ... | @@ -259,25 +259,19 @@ async def handle_create_report(
return fastapi.Response(status_code=500)
-@router.put("/{chat_id}/title")
-async def handle_update_title(
+@router.put("/{chat_id}/update") | inference/server/oasst_inference_server/routes/chats.py | 26 | Python | 0.071 | question | 30 | 42 | 42 | false | Unify title update and visibility update inference endpoints | 2,627 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | olliestanley | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/routes/chats.py
PR Title: Unify title update and visibility update inference endpoints
BEFORE CODE: try:
await ucr.update_score(message_id=message_id, score=vote_request.score)
return fastapi.Response... | Maybe "/{chat_id/}" is enough? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
if not task.collective:
task.done = True
self.db.add(task)
logger.debug(f"inserting TextLabels for {message_id=}, {text_labels_id=}")
model = TextLabels(
id=text_labels_id,
api_client_id=self.api_client.id,
message_id=messa... | send message_id inside send_new_report_message, this would be easier for the mod team to search the DB | if not task.collective:
task.done = True
self.db.add(task)
logger.debug(f"inserting TextLabels for {message_id=}, {text_labels_id=}")
model = TextLabels(
id=text_labels_id,
api_client_id=self.api_client.id,
message_id=messa... | @@ -546,6 +547,8 @@ def store_text_labels(self, text_labels: protocol_schema.TextLabels) -> tuple[Te
message_id, protocol_schema.EmojiOp.add, protocol_schema.EmojiCode.red_flag
)
+ discord.send_new_report_message(message=message, label_text=text_labels.... | backend/oasst_backend/prompt_repository.py | 26 | Python | 0.357 | suggestion | 102 | 51 | 51 | false | send flagged message notification to discord | 2,641 | LAION-AI/Open-Assistant | 29,137 | Python | melvinebenezer | notmd | Review this pull request change.
Language: Python
File: backend/oasst_backend/prompt_repository.py
PR Title: send flagged message notification to discord
BEFORE CODE: if not task.collective:
task.done = True
self.db.add(task)
logger.debug(f"inserting TextLabels f... | send message_id inside send_new_report_message, this would be easier for the mod team to search the DB | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import requests
from loguru import logger
from oasst_backend.config import settings
from oasst_backend.models.message import Message
ROOT_ENDPOINT = "https://discord.com/api/v10"
def send_new_report_message(
message: Message,
label_text: str,
) -> None:
if settings.DISCORD_API_KEY is None or settings.DIS... | maybe rename to `maybe_send_discord_report_message` | from uuid import UUID
import requests
from loguru import logger
from oasst_backend.config import settings
from oasst_backend.models.message import Message
ROOT_ENDPOINT = "https://discord.com/api/v10"
def send_new_report_message(message: Message, label_text: str, user_id: UUID) -> None:
if settings.DISCORD_API_... | @@ -0,0 +1,45 @@
+import requests
+from loguru import logger
+from oasst_backend.config import settings
+from oasst_backend.models.message import Message
+
+ROOT_ENDPOINT = "https://discord.com/api/v10"
+
+
+def send_new_report_message( | backend/oasst_backend/utils/discord.py | 9 | Python | 0.429 | suggestion | 51 | 34 | 34 | false | send flagged message notification to discord | 2,641 | LAION-AI/Open-Assistant | 29,137 | Python | yk | notmd | Review this pull request change.
Language: Python
File: backend/oasst_backend/utils/discord.py
PR Title: send flagged message notification to discord
BEFORE CODE: import requests
from loguru import logger
from oasst_backend.config import settings
from oasst_backend.models.message import Message
ROOT_ENDPOINT = "ht... | maybe rename to `maybe_send_discord_report_message` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
throw new Error("User not found");
}
const backendUser = convertToBackendUserCore(frontendUser);
if (backendUser.auth_method === "discord") {
const discordAccount = frontendUser.accounts.find((a) => a.provider === "discord");
discordAvatarRefresh.updateImageIfNeces... | the backend does a lazy upsert so we should be safe. see https://github.com/LAION-AI/Open-Assistant/blob/6ba240f590787b40d56b3d14e19fd38b9488c850/backend/oasst_backend/api/v1/frontend_users.py#L62 | if (!frontendUser) {
// should never reach here
throw new Error("User not found");
}
// TODO avoid duplicate logic with the signIn event
if (frontendUser.isNew && !frontendUser.name) {
// jwt callback is called before signIn event, so we need to assign the ... | @@ -222,9 +227,9 @@ export default function auth(req: NextApiRequest, res: NextApiResponse) {
* in the frontend, when the user accepts the tos, we do a full refresh
* which means this function will be called again.
*/
- const { user_id, tos_acceptance_date } = await ... | website/src/pages/api/auth/[...nextauth].ts | 26 | TypeScript | 0.429 | suggestion | 196 | 51 | 51 | false | create backend user when generate jwt | 2,702 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | notmd | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/auth/[...nextauth].ts
PR Title: create backend user when generate jwt
BEFORE CODE: throw new Error("User not found");
}
const backendUser = convertToBackendUserCore(frontendUser);
if (backendUser.aut... | the backend does a lazy upsert so we should be safe. see https://github.com/LAION-AI/Open-Assistant/blob/6ba240f590787b40d56b3d14e19fd38b9488c850/backend/oasst_backend/api/v1/frontend_users.py#L62 | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
};
export default function auth(req: NextApiRequest, res: NextApiResponse) {
return NextAuth(req, res, {
...authOptions,
callbacks: {
...authOptions.callbacks,
/**
* When creating a token, fetch the user's role and inject it in the token.
* This let's use forward the role to the ses... | we already create a username in line 143? | },
},
};
export default function auth(req: NextApiRequest, res: NextApiResponse) {
return NextAuth(req, res, {
...authOptions,
callbacks: {
...authOptions.callbacks,
/**
* When creating a token, fetch the user's role and inject it in the token.
* This let's use forward the rol... | @@ -200,6 +200,24 @@ export default function auth(req: NextApiRequest, res: NextApiResponse) {
select: { name: true, role: true, isNew: true, accounts: true, id: true },
});
+ if (!frontendUser) {
+ // should never reach here
+ throw new Error("User not found");
+ }... | website/src/pages/api/auth/[...nextauth].ts | 26 | TypeScript | 0.143 | question | 41 | 51 | 51 | false | create backend user when generate jwt | 2,702 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/auth/[...nextauth].ts
PR Title: create backend user when generate jwt
BEFORE CODE: };
export default function auth(req: NextApiRequest, res: NextApiResponse) {
return NextAuth(req, res, {
...authOptions,
callbacks: {
... | we already create a username in line 143? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import json
import os
import random
import pandas as pd
# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked in the README.
CURRENT_DIR = os.path.dirname(__file__) # Gets directory path of the current python module
df = pd.read_csv(os.path.join(CURRENT_DIR, ... | you can also do something like
`df["Title"].apply(lambda x: "".join(x.splitlines()).strip())`
but it's fine | import json
import os
import random
import kaggle
import pandas as pd
# Authenticate the Kaggle API client
kaggle.api.authenticate()
# Download and extract the dataset to the download_path directory
download_path = os.path.join(os.getcwd(), "data", "datasets", "poetry_instruction")
kaggle.api.dataset_download_files(... | @@ -0,0 +1,151 @@
+import json
+import os
+import random
+
+import pandas as pd
+
+# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked in the README.
+
+CURRENT_DIR = os.path.dirname(__file__) # Gets directory path of the current python module
+df = pd.read_cs... | data/datasets/poetry_instruction/prepare.py | 21 | Python | 0.5 | suggestion | 109 | 46 | 46 | false | add poetry dataset setup | 2,730 | LAION-AI/Open-Assistant | 29,137 | Python | sedthh | check3D | Review this pull request change.
Language: Python
File: data/datasets/poetry_instruction/prepare.py
PR Title: add poetry dataset setup
BEFORE CODE: import json
import os
import random
import pandas as pd
# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked... | you can also do something like
`df["Title"].apply(lambda x: "".join(x.splitlines()).strip())`
but it's fine | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
# Shuffle the rows in the DataFrame.
df = df.sample(frac=1)
# Split the DataFrame into two DataFrames, one for titling tasks and one for writing tasks.
writing_tasks = df.iloc[:num_writing_tasks]
titling_tasks = df.iloc[num_writing_tasks:]
prepared_data = []
# Loop through the writing tasks and process them.
for ind... | you can also do string.format() with "blabla {topic}" like https://www.w3schools.com/python/ref_string_format.asp
but this is okay too |
# Shuffling the dataset and delegating 5% to titling tasks.
# Calculating the number of titling tasks and writing tasks.
num_rows = len(df)
num_titling_tasks = int(num_rows * 0.05)
num_writing_tasks = num_rows - num_titling_tasks
# Shuffle the rows in the DataFrame.
df = df.sample(frac=1)
# Split the DataFrame into ... | @@ -0,0 +1,151 @@
+import json
+import os
+import random
+
+import pandas as pd
+
+# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked in the README.
+
+CURRENT_DIR = os.path.dirname(__file__) # Gets directory path of the current python module
+df = pd.read_cs... | data/datasets/poetry_instruction/prepare.py | 26 | Python | 0.357 | suggestion | 138 | 51 | 51 | false | add poetry dataset setup | 2,730 | LAION-AI/Open-Assistant | 29,137 | Python | sedthh | check3D | Review this pull request change.
Language: Python
File: data/datasets/poetry_instruction/prepare.py
PR Title: add poetry dataset setup
BEFORE CODE: # Shuffle the rows in the DataFrame.
df = df.sample(frac=1)
# Split the DataFrame into two DataFrames, one for titling tasks and one for writing tasks.
writing_tasks =... | you can also do string.format() with "blabla {topic}" like https://www.w3schools.com/python/ref_string_format.asp
but this is okay too | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
# Shuffle the rows in the DataFrame.
df = df.sample(frac=1)
# Split the DataFrame into two DataFrames, one for titling tasks and one for writing tasks.
writing_tasks = df.iloc[:num_writing_tasks]
titling_tasks = df.iloc[num_writing_tasks:]
prepared_data = []
# Loop through the writing tasks and process them.
for in... | please use `pd.isna(instruction)` or something similar, the "nan" is an artifact from converting NaN (missing/None) to string | ]
# Shuffling the dataset and delegating 5% to titling tasks.
# Calculating the number of titling tasks and writing tasks.
num_rows = len(df)
num_titling_tasks = int(num_rows * 0.05)
num_writing_tasks = num_rows - num_titling_tasks
# Shuffle the rows in the DataFrame.
df = df.sample(frac=1)
# Split the DataFrame int... | @@ -0,0 +1,151 @@
+import json
+import os
+import random
+
+import pandas as pd
+
+# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked in the README.
+
+CURRENT_DIR = os.path.dirname(__file__) # Gets directory path of the current python module
+df = pd.read_cs... | data/datasets/poetry_instruction/prepare.py | 26 | Python | 0.571 | suggestion | 125 | 51 | 51 | false | add poetry dataset setup | 2,730 | LAION-AI/Open-Assistant | 29,137 | Python | sedthh | check3D | Review this pull request change.
Language: Python
File: data/datasets/poetry_instruction/prepare.py
PR Title: add poetry dataset setup
BEFORE CODE:
# Shuffle the rows in the DataFrame.
df = df.sample(frac=1)
# Split the DataFrame into two DataFrames, one for titling tasks and one for writing tasks.
writing_tasks ... | please use `pd.isna(instruction)` or something similar, the "nan" is an artifact from converting NaN (missing/None) to string | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import json
import os
import random
import pandas as pd
# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked in the README.
CURRENT_DIR = os.path.dirname(__file__) # Gets directory path of the current python module
df = pd.read_csv(os.path.join(CURRENT_DIR, ... | mabye you can also add a KAGGLE dataset loader to unpack this (I can give you an example code for this) | import json
import os
import random
import kaggle
import pandas as pd
# Authenticate the Kaggle API client
kaggle.api.authenticate()
# Download and extract the dataset to the download_path directory
download_path = os.path.join(os.getcwd(), "data", "datasets", "poetry_instruction")
kaggle.api.dataset_download_files(... | @@ -0,0 +1,151 @@
+import json
+import os
+import random
+
+import pandas as pd
+
+# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked in the README.
+
+CURRENT_DIR = os.path.dirname(__file__) # Gets directory path of the current python module | data/datasets/poetry_instruction/prepare.py | 9 | Python | 0.357 | suggestion | 103 | 34 | 34 | false | add poetry dataset setup | 2,730 | LAION-AI/Open-Assistant | 29,137 | Python | sedthh | check3D | Review this pull request change.
Language: Python
File: data/datasets/poetry_instruction/prepare.py
PR Title: add poetry dataset setup
BEFORE CODE: import json
import os
import random
import pandas as pd
# Read the CSV file into a pandas DataFrame. The CSV file was downloaded from Kaggle, and the source is linked... | mabye you can also add a KAGGLE dataset loader to unpack this (I can give you an example code for this) | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
router = fastapi.APIRouter(
prefix="/chats",
tags=["chats"],
)
@router.get("")
async def list_chats(
include_hidden: bool = False,
ucr: UserChatRepository = Depends(deps.create_user_chat_repository),
) -> chat_schema.ListChatsResponse:
"""Lists all chats."""
logger.info("Listing all chats.")
... | why do we base64 encode / decode the ids? | from sse_starlette.sse import EventSourceResponse
router = fastapi.APIRouter(
prefix="/chats",
tags=["chats"],
)
@router.get("")
async def list_chats(
include_hidden: bool = False,
ucr: UserChatRepository = Depends(deps.create_user_chat_repository),
limit: int | None = Query(10, gt=0, le=100),
... | @@ -21,12 +22,47 @@
async def list_chats(
include_hidden: bool = False,
ucr: UserChatRepository = Depends(deps.create_user_chat_repository),
+ limit: int | None = Query(10, gt=0, le=100),
+ after: str | None = None,
+ before: str | None = None,
) -> chat_schema.ListChatsResponse:
"""Lists all ... | inference/server/oasst_inference_server/routes/chats.py | 26 | Python | 0.071 | question | 41 | 51 | 51 | false | support pagination for chat list | 2,773 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: Python
File: inference/server/oasst_inference_server/routes/chats.py
PR Title: support pagination for chat list
BEFORE CODE:
router = fastapi.APIRouter(
prefix="/chats",
tags=["chats"],
)
@router.get("")
async def list_chats(
include_hidden: bool = False,
... | why do we base64 encode / decode the ids? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
from enum import Enum, unique
class Mode(Enum):
sft = "sft"
rm = "rm"
rl = "rl"
@unique
class Language(str, Enum):
AB = "ab" # Abkhazian
AA = "aa" # Afar
AF = "af" # Afrikaans
AK = "ak" # Akan
SQ = "sq" # Albanian
AM = "am" # Amharic
AR = "ar" # Arabic
AN = "an" #... | Are these all BCP 47 lang-codes? Should document which standard is used, the human-demonstration collection backend uses BCP 47. We might also think about using a lib like for example https://pypi.org/project/langcodes/ ? | from enum import Enum
class Mode(Enum):
sft = "sft"
rm = "rm"
rl = "rl"
| @@ -0,0 +1,195 @@
+from enum import Enum, unique
+
+
+class Mode(Enum):
+ sft = "sft"
+ rm = "rm"
+ rl = "rl"
+
+
+@unique
+class Language(str, Enum):
+ AB = "ab" # Abkhazian
+ AA = "aa" # Afar
+ AF = "af" # Afrikaans
+ AK = "ak" # Akan
+ SQ = "sq" # Albanian
+ AM = "am" # Amharic | model/model_training/custom_datasets/entities.py | 17 | Python | 0.5 | question | 222 | 42 | 8 | false | update entities | 2,809 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | CloseChoice | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/entities.py
PR Title: update entities
BEFORE CODE: from enum import Enum, unique
class Mode(Enum):
sft = "sft"
rm = "rm"
rl = "rl"
@unique
class Language(str, Enum):
AB = "ab" # Abkhazian
AA = "aa" ... | Are these all BCP 47 lang-codes? Should document which standard is used, the human-demonstration collection backend uses BCP 47. We might also think about using a lib like for example https://pypi.org/project/langcodes/ ? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
ActionRowBuilder,
ButtonStyle,
ButtonBuilder,
} from "discord.js";
import { createInferenceClient } from "../modules/inference/client.js";
import redis from "../modules/redis.js";
export default {
disablePing: null,
data: new SlashCommandBuilder()
.setName("chat")
.setDescription("Chat with an Open A... | It would be best if the list of available models in the future would be requested at the inference-server (/models endpoint). It will be constantly changing .. you were probably irritated that so many are displayed in the staging environment .. but that's really only a "problem" in staging. | ActionRowBuilder,
ButtonStyle,
ButtonBuilder,
} from "discord.js";
import redis from "../modules/redis.js";
import chatFN from "../modules/chat.js";
export default {
disablePing: null,
data: new SlashCommandBuilder()
.setName("chat")
.setDescription("Chat with an Open Assistant AI.")
.addStringOp... | @@ -23,10 +23,16 @@ export default {
.setName("model")
.setDescription("The model you want to use for the AI.")
.setRequired(false)
- .addChoices({
- name: "OA_SFT_Llama_30B",
- value: "OA_SFT_Llama_30B",
- })
+ .addChoices(
+ {
+ nam... | discord-bots/oa-bot-js/src/commands/chat.ts | 26 | TypeScript | 0.429 | suggestion | 291 | 51 | 51 | false | Discord bot: Pythia model | 2,831 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | mrloldev | Review this pull request change.
Language: TypeScript
File: discord-bots/oa-bot-js/src/commands/chat.ts
PR Title: Discord bot: Pythia model
BEFORE CODE: ActionRowBuilder,
ButtonStyle,
ButtonBuilder,
} from "discord.js";
import { createInferenceClient } from "../modules/inference/client.js";
import redis from ... | It would be best if the list of available models in the future would be requested at the inference-server (/models endpoint). It will be constantly changing .. you were probably irritated that so many are displayed in the staging environment .. but that's really only a "problem" in staging. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
setMessages((messages) => [...messages, dummyMessage]);
prompter_message = await post(API_ROUTES.CREATE_PROMPTER_MESSAGE, { arg: prompter_arg });
// filter out the dummy message and push the real one
setMessages((messages) => [...messages.filter((m) => m.id !== "__dummy__"), prompter_mes... | This is a magic number, I tested on different screen widths and found this is the most optimal one lol | setMessages((messages) => [...messages, dummyMessage]);
prompter_message = await post(API_ROUTES.CREATE_PROMPTER_MESSAGE, { arg: prompter_arg });
// filter out the dummy message and push the real one
setMessages((messages) => [...messages.filter((m) => m.id !== "__dummy__"), prompter_mes... | @@ -200,6 +200,22 @@ export const ChatConversation = memo(function ChatConversation({ chatId, getConf
[createAndFetchAssistantMessage, isSending, setIsSending]
);
+ const messagesEndRef = useRef<HTMLDivElement>(null);
+ const chatContainerRef = useRef<HTMLDivElement>(null);
+
+ useEffect(() => {
+ if (!... | website/src/components/Chat/ChatConversation.tsx | 26 | TypeScript | 0.357 | suggestion | 102 | 51 | 51 | false | Implement auto scroll when stream message | 2,839 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | notmd | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConversation.tsx
PR Title: Implement auto scroll when stream message
BEFORE CODE: setMessages((messages) => [...messages, dummyMessage]);
prompter_message = await post(API_ROUTES.CREATE_PROMPTER_MESSAGE, { ar... | This is a magic number, I tested on different screen widths and found this is the most optimal one lol | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { useEffect } from "react";
import { useFormContext } from "react-hook-form";
import { ChatConfigFormData } from "src/types/Chat";
import { setConfigCache } from "src/utils/chat";
export const ChatConfigSaver = () => {
const { watch } = useFormContext<ChatConfigFormData>();
const config = watch();
useEffe... | when we change presets, we use the function `reset`, which sets the `isDirty` to false, so we are not saving. | import { useEffect } from "react";
import { useFormContext } from "react-hook-form";
import { ChatConfigFormData } from "src/types/Chat";
import { setConfigCache } from "src/utils/chat";
export const ChatConfigSaver = () => {
const { watch, formState } = useFormContext<ChatConfigFormData>();
const config = watch()... | @@ -4,14 +4,11 @@ import { ChatConfigFormData } from "src/types/Chat";
import { setConfigCache } from "src/utils/chat";
export const ChatConfigSaver = () => {
- const { watch, formState } = useFormContext<ChatConfigFormData>();
+ const { watch } = useFormContext<ChatConfigFormData>();
const config = watch();
... | website/src/components/Chat/ChatConfigSaver.tsx | 11 | TypeScript | 0.571 | suggestion | 109 | 15 | 17 | false | Fix saving model between chats | 2,843 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | AbdBarho | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConfigSaver.tsx
PR Title: Fix saving model between chats
BEFORE CODE: import { useEffect } from "react";
import { useFormContext } from "react-hook-form";
import { ChatConfigFormData } from "src/types/Chat";
import { setConf... | when we change presets, we use the function `reset`, which sets the `isDirty` to false, so we are not saving. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
return pydantic.parse_obj_as(Configuration, json_data)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--device", default="cuda", type=str, help="device to use")
parser.add_argument("--device-index", default=0, type=int, help="device index")
parser.add_argument("--model-... | Can we call this `--int8`? Could be confusing as-is, as `--half` is also quantisation |
return pydantic.parse_obj_as(Configuration, json_data)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--device", default="cuda", type=str, help="device to use")
parser.add_argument("--device-index", default=0, type=int, help="device index")
parser.add_argument("--model-... | @@ -226,6 +226,7 @@ def parse_args():
parser.add_argument("--num-samples", type=int, default=2, help="number of sampling runs per configuration")
parser.add_argument("--config", type=str, default="config/default.json", help="configuration file path")
parser.add_argument("--half", action="store_true", def... | model/model_eval/manual/sampling_report.py | 26 | Python | 0.429 | question | 85 | 51 | 51 | false | Add quantization/8bit model loading support for sampling_report.py | 2,857 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | toiletpapercode | Review this pull request change.
Language: Python
File: model/model_eval/manual/sampling_report.py
PR Title: Add quantization/8bit model loading support for sampling_report.py
BEFORE CODE:
return pydantic.parse_obj_as(Configuration, json_data)
def parse_args():
parser = argparse.ArgumentParser()
pars... | Can we call this `--int8`? Could be confusing as-is, as `--half` is also quantisation | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { withoutRole } from "src/lib/auth";
import { isChatEnable } from "src/lib/isChatEnable";
import { createInferenceClient } from "src/lib/oasst_inference_client";
const handler = withoutRole("banned", async (req, res, token) => {
if (!isChatEnable()) {
return res.status(404).end();
}
const client = cre... | Can you remove this option and hard code it to this file? | import { withoutRole } from "src/lib/auth";
import { isChatEnable } from "src/lib/isChatEnable";
import { createInferenceClient } from "src/lib/oasst_inference_client";
import { GetChatsParams } from "src/types/Chat";
export const PAGE_SIZE = 20;
const handler = withoutRole("banned", async (req, res, token) => {
if... | @@ -13,7 +13,17 @@ const handler = withoutRole("banned", async (req, res, token) => {
if (req.query.chat_id) {
data = await client.get_chat(req.query.chat_id as string);
} else {
- data = await client.get_my_chats();
+ const params = {};
+ if (req.query.limit) { | website/src/pages/api/chat/index.ts | 17 | TypeScript | 0.286 | refactor | 57 | 42 | 42 | false | feat(website): support pagination for chat list | 2,860 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | mtosity | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/chat/index.ts
PR Title: feat(website): support pagination for chat list
BEFORE CODE: import { withoutRole } from "src/lib/auth";
import { isChatEnable } from "src/lib/isChatEnable";
import { createInferenceClient } from "src/lib/oasst... | Can you remove this option and hard code it to this file? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { withoutRole } from "src/lib/auth";
import { isChatEnable } from "src/lib/isChatEnable";
import { createInferenceClient } from "src/lib/oasst_inference_client";
export const PAGE_SIZE = 20;
const handler = withoutRole("banned", async (req, res, token) => {
if (!isChatEnable()) {
return res.status(404).e... | please add type `params: GetChatParams` just to be sure.
Otherwise looks good! | import { withoutRole } from "src/lib/auth";
import { isChatEnable } from "src/lib/isChatEnable";
import { createInferenceClient } from "src/lib/oasst_inference_client";
import { GetChatsParams } from "src/types/Chat";
export const PAGE_SIZE = 20;
const handler = withoutRole("banned", async (req, res, token) => {
if... | @@ -13,7 +15,16 @@ const handler = withoutRole("banned", async (req, res, token) => {
if (req.query.chat_id) {
data = await client.get_chat(req.query.chat_id as string);
} else {
- data = await client.get_my_chats();
+ const params = { | website/src/pages/api/chat/index.ts | 18 | TypeScript | 0.5 | suggestion | 81 | 43 | 43 | false | feat(website): support pagination for chat list | 2,860 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | mtosity | Review this pull request change.
Language: TypeScript
File: website/src/pages/api/chat/index.ts
PR Title: feat(website): support pagination for chat list
BEFORE CODE: import { withoutRole } from "src/lib/auth";
import { isChatEnable } from "src/lib/isChatEnable";
import { createInferenceClient } from "src/lib/oasst... | please add type `params: GetChatParams` just to be sure.
Otherwise looks good! | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
</>
);
};
type WithSpanCell<T> = Cell<T, unknown> & { span?: number };
const DataTableRow = <T,>({ row }: { row: Row<T> }) => {
const cells: WithSpanCell<T>[] = row.getVisibleCells();
const renderCells: WithSpanCell<T>[] = [];
for (let i = 0; i < cells.length; i++) {
const cell = cells[i];
const ... | We dont want to add this bahavior on the DataTable component | </>
);
};
type WithSpanCell<T> = Cell<T, unknown> & { span?: number };
const DataTableRow = <T,>({ row }: { row: Row<T> }) => {
const cells: WithSpanCell<T>[] = row.getVisibleCells();
const renderCells: WithSpanCell<T>[] = [];
for (let i = 0; i < cells.length; i++) {
const cell = cells[i];
const ... | @@ -199,7 +199,7 @@ const DataTableRow = <T,>({ row }: { row: Row<T> }) => {
{renderCells.map((cell) => {
const props = cell.column.columnDef.meta?.cellProps?.(cell);
return (
- <Td key={cell.id} {...props} colSpan={cell.span}>
+ <Td key={cell.id} {...props} colSpan={cell.span... | website/src/components/DataTable/DataTable.tsx | 26 | TypeScript | 0.214 | suggestion | 60 | 51 | 51 | false | Fix display name spill-over (#2851) | 2,863 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | scorbett123 | Review this pull request change.
Language: TypeScript
File: website/src/components/DataTable/DataTable.tsx
PR Title: Fix display name spill-over (#2851)
BEFORE CODE: </>
);
};
type WithSpanCell<T> = Cell<T, unknown> & { span?: number };
const DataTableRow = <T,>({ row }: { row: Row<T> }) => {
const cells:... | We dont want to add this bahavior on the DataTable component | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
class JokeExplaination(Dataset):
name = "joke"
url = "https://gist.github.com/theblackcat102/42b697e24a13fdb499e20edfbf618361/raw/1834dca207898c15f93b809d1195f6f6e47c9e1e/joke_explained.jsonl"
def __init__(self, cache_dir) -> None:
super().__init__()
os.makedirs(cache_dir, exist_ok=True)
... | you rely here on the order of the keywords. Could we make this explicit by using
```suggestion
self.pairs.append(DatasetEntry(questions=[joke], answers=[explanation]))
``` |
class JokeExplaination(Dataset):
name = "joke"
url = "https://gist.github.com/theblackcat102/42b697e24a13fdb499e20edfbf618361/raw/1834dca207898c15f93b809d1195f6f6e47c9e1e/joke_explained.jsonl"
def __init__(self, cache_dir) -> None:
super().__init__()
os.makedirs(cache_dir, exist_ok=True)
... | @@ -343,16 +343,16 @@ def __init__(self, cache_dir) -> None:
# DO NOT change this
# its the data that had syntax error
explanation = data["explaination"]
- self.pairs.append((joke, explanation))
+ self.pairs.append(DatasetEntry([joke], [ex... | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.786 | suggestion | 190 | 51 | 51 | false | Joke explanation generalization | 2,899 | LAION-AI/Open-Assistant | 29,137 | Python | CloseChoice | sampatkalyan | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: Joke explanation generalization
BEFORE CODE:
class JokeExplaination(Dataset):
name = "joke"
url = "https://gist.github.com/theblackcat102/42b697e24a13fdb499e20edfbf618361/raw/1834dca207898c... | you rely here on the order of the keywords. Could we make this explicit by using
```suggestion
self.pairs.append(DatasetEntry(questions=[joke], answers=[explanation]))
``` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
url = "https://gist.github.com/theblackcat102/42b697e24a13fdb499e20edfbf618361/raw/1834dca207898c15f93b809d1195f6f6e47c9e1e/joke_explained.jsonl"
def __init__(self, cache_dir) -> None:
super().__init__()
os.makedirs(cache_dir, exist_ok=True)
joke_explain_filename = os.path.join(cache_di... | You mentioned already that question and answer are not use, so please remove them and also this `if` statement. | url = "https://gist.github.com/theblackcat102/42b697e24a13fdb499e20edfbf618361/raw/1834dca207898c15f93b809d1195f6f6e47c9e1e/joke_explained.jsonl"
def __init__(self, cache_dir) -> None:
super().__init__()
os.makedirs(cache_dir, exist_ok=True)
joke_explain_filename = os.path.join(cache_di... | @@ -343,16 +343,16 @@ def __init__(self, cache_dir) -> None:
# DO NOT change this
# its the data that had syntax error
explanation = data["explaination"]
- self.pairs.append((joke, explanation))
+ self.pairs.append(DatasetEntry([joke], [ex... | model/model_training/custom_datasets/qa_datasets.py | 26 | Python | 0.571 | suggestion | 111 | 51 | 51 | false | Joke explanation generalization | 2,899 | LAION-AI/Open-Assistant | 29,137 | Python | CloseChoice | sampatkalyan | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/qa_datasets.py
PR Title: Joke explanation generalization
BEFORE CODE: url = "https://gist.github.com/theblackcat102/42b697e24a13fdb499e20edfbf618361/raw/1834dca207898c15f93b809d1195f6f6e47c9e1e/joke_explained.jsonl"
... | You mentioned already that question and answer are not use, so please remove them and also this `if` statement. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { Box } from "@chakra-ui/react";
export const ChatWarning = () => {
return (
<Box fontSize="sm" textAlign="center" mb="2" className="max-w-5xl mx-auto">
This Assistant is a demonstration version that does not have internet access. It may generate incorrect or
misleading information. It is not ... | can we put this back into the locale? we want it to be translated. | import { Box } from "@chakra-ui/react";
import { useTranslation } from "next-i18next";
export const ChatWarning = () => {
const { t } = useTranslation("chat");
return (
<Box fontSize="sm" textAlign="center" mb="2" className="max-w-5xl mx-auto">
{t("warning")}
</Box>
);
};
| @@ -1,27 +1,10 @@
-import { Flex } from "@chakra-ui/react";
-import { Fragment } from "react";
-import { TeamMember } from "src/components/TeamMember";
-
-import data from "../../data/warning.json";
-
-const warn = ["warning"] as const;
-
-const rootProps = {
- bg: "transparent",
- borderRadius: 0,
-};
+import { Box ... | website/src/components/Chat/ChatWarning.tsx | 7 | TypeScript | 0.214 | question | 66 | 11 | 12 | false | update chat warning design | 2,924 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | notmd | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatWarning.tsx
PR Title: update chat warning design
BEFORE CODE: import { Box } from "@chakra-ui/react";
export const ChatWarning = () => {
return (
<Box fontSize="sm" textAlign="center" mb="2" className="max-w-5xl mx-au... | can we put this back into the locale? we want it to be translated. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
def __getitem__(self, index):
return self.data[index]
def load_oasst_export(
input_file_path: str | Path,
hf_dataset_name: str,
val_split: float = 0.2,
lang: str = "en",
top_k: Optional[int] = None,
manual_seed: int = 287631038922,
data_path: str | Path = None,
use_hf_datas... | The format between trees and messages in the HF dataset differs greatly .. some conversion is required to restore the OA format. |
def __getitem__(self, index):
return self.data[index]
def load_oasst_export(
input_file_path: Optional[str | Path] = None,
hf_dataset_name: Optional[str] = "OpenAssistant/oasst1",
val_split: float = 0.2,
lang: str = "en",
top_k: Optional[int] = None,
manual_seed: int = 28763103892... | @@ -35,6 +38,10 @@ def load_oasst_export(
generator = Generator()
generator.manual_seed(manual_seed)
+ if use_hf_dataset and hf_dataset_name:
+ hf_dataset = load_dataset(hf_dataset_name)
+ return hf_dataset["train"], hf_dataset["validation"] | model/model_training/custom_datasets/oasst_dataset.py | 26 | Python | 0.357 | style | 128 | 51 | 51 | false | Add data loader for HF oasst1 | 2,951 | LAION-AI/Open-Assistant | 29,137 | Python | andreaskoepf | grgau | Review this pull request change.
Language: Python
File: model/model_training/custom_datasets/oasst_dataset.py
PR Title: Add data loader for HF oasst1
BEFORE CODE: def __getitem__(self, index):
return self.data[index]
def load_oasst_export(
input_file_path: str | Path,
hf_dataset_name: str,
... | The format between trees and messages in the HF dataset differs greatly .. some conversion is required to restore the OA format. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
from __future__ import annotations
from datetime import datetime
from typing import Literal, Optional
from pydantic import BaseModel, conint
class LabelAvgValue(BaseModel):
value: float | None
count: int
LabelValues = dict
class ExportMessageEvent(BaseModel):
type: str
user_id: str | None
clas... | What is the reason for this change (and the similar one for `emojis` below)? | from __future__ import annotations
from datetime import datetime
from typing import Literal, Optional
from pydantic import BaseModel, conint
class LabelAvgValue(BaseModel):
value: float | None
count: int
LabelValues = dict[str, LabelAvgValue]
class ExportMessageEvent(BaseModel):
type: str
user_i... | @@ -11,7 +11,7 @@ class LabelAvgValue(BaseModel):
count: int
-LabelValues = dict[str, LabelAvgValue] | oasst-data/oasst_data/schemas.py | 14 | Python | 0.429 | question | 76 | 39 | 39 | false | Add data loader for HF oasst1 | 2,951 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | grgau | Review this pull request change.
Language: Python
File: oasst-data/oasst_data/schemas.py
PR Title: Add data loader for HF oasst1
BEFORE CODE: from __future__ import annotations
from datetime import datetime
from typing import Literal, Optional
from pydantic import BaseModel, conint
class LabelAvgValue(BaseModel... | What is the reason for this change (and the similar one for `emojis` below)? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
onError: () => {
toast({
title: "Failed to load chat",
status: "error",
});
},
});
const createAndFetchAssistantMessage = useCallback(
async ({ parentId, chatId }: { parentId: string; chatId: string }) => {
const { model_config_name, ...sampling_parameters } = getConfi... | This always is `OasstError` since we already transformed the error, the check to make TS happy | },
onError: () => {
toast({
title: "Failed to load chat",
status: "error",
});
},
});
const createAndFetchAssistantMessage = useCallback(
async ({ parentId, chatId }: { parentId: string; chatId: string }) => {
const { model_config_name, ...sampling_parameters } = g... | @@ -58,9 +59,21 @@ export const ChatConversation = memo(function ChatConversation({ chatId, getConf
sampling_parameters,
};
- const assistant_message: InferenceMessage = await post(API_ROUTES.CREATE_ASSISTANT_MESSAGE, {
- arg: assistant_arg,
- });
+ let assistant_message: Inferen... | website/src/components/Chat/ChatConversation.tsx | 26 | TypeScript | 0.5 | suggestion | 94 | 51 | 51 | false | Handle chat error | 2,959 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | notmd | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatConversation.tsx
PR Title: Handle chat error
BEFORE CODE: onError: () => {
toast({
title: "Failed to load chat",
status: "error",
});
},
});
const createAndFetchAssistantMessage = use... | This always is `OasstError` since we already transformed the error, the check to make TS happy | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
Stack,
Textarea,
useDisclosure,
} from "@chakra-ui/react";
import { Cog, Edit, Plus } from "lucide-react";
import { AttachmentIcon, WarningIcon, CheckCircleIcon, CloseIcon } from "@chakra-ui/icons";
import { ChangeEvent, useCallback, useMemo, useEffect, useState, useRef } from "react";
import { useTranslation } f... | Since you alreay has a fall back `|| []`, I don't think we need `?` and `?? - 1` here anymore
```suggestion
return plugins.findIndex((plugin) => plugin.enabled);
``` | Stack,
Textarea,
useDisclosure,
} from "@chakra-ui/react";
import { Cog, Edit, Plus } from "lucide-react";
import { AttachmentIcon, WarningIcon, CheckCircleIcon, CloseIcon } from "@chakra-ui/icons";
import { ChangeEvent, useCallback, useMemo, useEffect, useState, useRef } from "react";
import { useTranslation } f... | @@ -56,7 +49,7 @@ export const PluginsChooser = ({ plugins }: { plugins: PluginEntry[] }) => {
const containerRef = useRef<HTMLDivElement | null>(null);
const activePluginIndex = useMemo(() => {
- return plugins.findIndex((plugin) => plugin.enabled);
+ return plugins?.findIndex((plugin) => plugin.enabled)... | website/src/components/Chat/PluginsChooser.tsx | 26 | TypeScript | 0.714 | suggestion | 172 | 51 | 51 | false | Fixed PluginChooser button width problem | 3,014 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | draganjovanovich | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/PluginsChooser.tsx
PR Title: Fixed PluginChooser button width problem
BEFORE CODE: Stack,
Textarea,
useDisclosure,
} from "@chakra-ui/react";
import { Cog, Edit, Plus } from "lucide-react";
import { AttachmentIcon, Warning... | Since you alreay has a fall back `|| []`, I don't think we need `?` and `?? - 1` here anymore
```suggestion
return plugins.findIndex((plugin) => plugin.enabled);
``` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
const plugin: PluginEntry = await post(API_ROUTES.GET_PLUGIN_CONFIG, {
arg: { plugin: newPlugins[selectedForEditPluginIndex!] },
});
// disable all existing plugins and enable new one
newPlugins.forEach((plugin) => (plugin.enabled = false));
newPlugins[selectedForEditPluginIndex!].... | ```suggestion
{activePluginIndex > -1 && plugins[activePluginIndex]?.enabled ? (
``` | const plugin: PluginEntry = await post(API_ROUTES.GET_PLUGIN_CONFIG, {
arg: { plugin: newPlugins[selectedForEditPluginIndex!] },
});
// disable all existing plugins and enable new one
newPlugins.forEach((plugin) => (plugin.enabled = false));
newPlugins[selectedForEditPluginIndex!].... | @@ -114,30 +107,30 @@ export const PluginsChooser = ({ plugins }: { plugins: PluginEntry[] }) => {
}, [setValue, plugins, handlePluginEdit]);
return (
- <FormControl mb="40px" ref={containerRef}>
- <Box position="fixed" zIndex="1">
- <Menu placement="auto">
- <MenuButton as={Button} pr="... | website/src/components/Chat/PluginsChooser.tsx | 26 | TypeScript | 0.571 | suggestion | 96 | 51 | 51 | false | Fixed PluginChooser button width problem | 3,014 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | draganjovanovich | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/PluginsChooser.tsx
PR Title: Fixed PluginChooser button width problem
BEFORE CODE: const plugin: PluginEntry = await post(API_ROUTES.GET_PLUGIN_CONFIG, {
arg: { plugin: newPlugins[selectedForEditPluginIndex!] },
... | ```suggestion
{activePluginIndex > -1 && plugins[activePluginIndex]?.enabled ? (
``` | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import json
import re
import threading
import requests
import transformers
from chat_chain_prompts import INSTRUCTIONS, OBSERVATION_SEQ, TOOLS_PREFIX, V2_ASST_PREFIX, V2_PROMPTER_PREFIX
from hf_langchain_inference import HFInference
from langchain.agents import Tool
from langchain.memory import ConversationBufferMemor... | Interesting! Why wasn't this detected during use and importing of the module? Does the current design fail quietly? | import json
import re
import requests
import transformers
from chat_chain_prompts import INSTRUCTIONS, OBSERVATION_SEQ, TOOLS_PREFIX, V2_ASST_PREFIX, V2_PROMPTER_PREFIX
from hf_langchain_inference import HFInference
from langchain.agents import Tool
from langchain.memory import ConversationBufferMemory
from langchain.... | @@ -11,7 +11,7 @@
from langchain.prompts import PromptTemplate
from loguru import logger
from oasst_shared.schemas import inference
-from opeanapi_parser import prepare_plugin_for_llm
+from openapi_parser import prepare_plugin_for_llm | inference/worker/chat_chain_utils.py | 13 | Python | 0.429 | bug | 115 | 38 | 38 | false | Fix plugins | 3,017 | LAION-AI/Open-Assistant | 29,137 | Python | pevogam | olliestanley | Review this pull request change.
Language: Python
File: inference/worker/chat_chain_utils.py
PR Title: Fix plugins
BEFORE CODE: import json
import re
import threading
import requests
import transformers
from chat_chain_prompts import INSTRUCTIONS, OBSERVATION_SEQ, TOOLS_PREFIX, V2_ASST_PREFIX, V2_PROMPTER_PREFIX
f... | Interesting! Why wasn't this detected during use and importing of the module? Does the current design fail quietly? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
<CreateChatButton
py="5"
leftIcon={<Plus size="16px"></Plus>}
variant="outline"
justifyContent="start"
colorScheme="blue"
borderRadius="lg"
mx="3"
mb="2"
onUpdated={handleCreateChat}
>
{t("create_chat")}
</CreateChatButton... | We still need to redirect user to another page after deleting, maybe latest chat or /chat page | <CreateChatButton
py="5"
leftIcon={<Plus size="16px"></Plus>}
variant="outline"
justifyContent="start"
colorScheme="blue"
borderRadius="lg"
mx="3"
mb="2"
onUpdated={handleCreateChat}
>
{t("create_chat")}
</CreateChatButton... | @@ -98,7 +98,13 @@ export const ChatListBase = memo(function ChatListBase({
}}
>
{chats.map((chat) => (
- <ChatListItem key={chat.id} chat={chat} onUpdateTitle={handleUpdateTitle} onHide={handleHide} />
+ <ChatListItem
+ key={chat.id}
+ chat={chat}
+ ... | website/src/components/Chat/ChatListBase.tsx | 26 | TypeScript | 0.286 | suggestion | 94 | 29 | 35 | false | add delete function | 3,019 | LAION-AI/Open-Assistant | 29,137 | Python | notmd | ruoniw | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/ChatListBase.tsx
PR Title: add delete function
BEFORE CODE: <CreateChatButton
py="5"
leftIcon={<Plus size="16px"></Plus>}
variant="outline"
justifyContent="start"
colorScheme="blue"
... | We still need to redirect user to another page after deleting, maybe latest chat or /chat page | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
yield token
else:
break
def finish(self, reason: Literal["length", "eos_token", "stop_sequence"]) -> Iterable[interface.Token]:
if reason == "stop_sequence":
end_sequence = ""
end_tokens = []
while self.tokens:
... | I think this function name is slightly misleading | yield token
else:
break
def finish(self, reason: Literal["length", "eos_token", "stop_sequence"]) -> Iterable[interface.Token]:
if reason == "stop_sequence":
end_sequence = ""
end_tokens = []
while self.tokens:
... | @@ -64,6 +64,19 @@ def finish(self, reason: Literal["length", "eos_token", "stop_sequence"]) -> Ite
yield from self.tokens
+def add_prefix_to_prompt(prompt: str): | inference/worker/utils.py | 26 | Python | 0.071 | suggestion | 49 | 51 | 51 | false | Cleaned up backend plugin code: Part 1 | 3,056 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | SerenityCodes | Review this pull request change.
Language: Python
File: inference/worker/utils.py
PR Title: Cleaned up backend plugin code: Part 1
BEFORE CODE: yield token
else:
break
def finish(self, reason: Literal["length", "eos_token", "stop_sequence"]) -> Iterable[interface.Tok... | I think this function name is slightly misleading | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
def get_max_input_length(worker_config: inference.WorkerConfig, plugin_used: bool):
max_input_length = worker_config.model_config.max_input_length
if plugin_used:
max_input_length = max_input_length - 1
return max_input_length
def truncate_prompt(
tokenizer: transformers.PreTrainedTokenizer,
... | As pointed out by Yannic in #3061, this length will turn out to be calculated wrong if the prompter prefix is added above. We should adjust this |
def truncate_prompt(
tokenizer: transformers.PreTrainedTokenizer,
worker_config: inference.WorkerConfig,
parameters: interface.GenerateStreamParameters,
prompt: str,
plugin_used: bool,
):
with shared_tokenizer_lock:
ids = tokenizer.encode(prompt)
max_input_length = get_max_input_le... | @@ -74,28 +87,22 @@ def truncate_prompt(
with shared_tokenizer_lock:
ids = tokenizer.encode(prompt)
- max_input_length = worker_config.model_config.max_input_length
-
- # make room for prompter prefix
- if plugin_used:
- max_input_length = max_input_length - 1
+ max_input_length = get... | inference/worker/utils.py | 26 | Python | 0.429 | bug | 144 | 51 | 51 | false | Cleaned up backend plugin code: Part 1 | 3,056 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | SerenityCodes | Review this pull request change.
Language: Python
File: inference/worker/utils.py
PR Title: Cleaned up backend plugin code: Part 1
BEFORE CODE: def get_max_input_length(worker_config: inference.WorkerConfig, plugin_used: bool):
max_input_length = worker_config.model_config.max_input_length
if plugin_used:
... | As pointed out by Yannic in #3061, this length will turn out to be calculated wrong if the prompter prefix is added above. We should adjust this | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
ids = ids[-(max_input_length - 1) :]
with shared_tokenizer_lock:
prompt = tokenizer.decode(ids)
prompt = add_prefix_to_prompt(prompt)
input_length = len(ids)
spare = max_total_tokens - input_length - 1
if not parameters.max_new_tokens:
parameters.max_new_toke... | I think this one should remain `warning`, it's relevant outside of debugging. | ids = tokenizer.encode(V2_PROMPTER_PREFIX) + ids
max_total_tokens = worker_config.model_config.max_total_length
input_length = len(ids)
spare = max_total_tokens - input_length - 1
if not parameters.max_new_tokens:
parameters.max_new_tokens = spare
elif parameters.max_new_to... | @@ -110,7 +117,7 @@ def wait_for_inference_server(http: "HttpClient", timeout: int = 600):
if time.time() > time_limit:
raise
sleep_duration = random.uniform(0, 10)
- logger.warning(f"Inference server not ready. Retrying in {sleep_duration:.2f} seconds")
+ ... | inference/worker/utils.py | 26 | Python | 0.571 | bug | 77 | 51 | 51 | false | Cleaned up backend plugin code: Part 1 | 3,056 | LAION-AI/Open-Assistant | 29,137 | Python | olliestanley | SerenityCodes | Review this pull request change.
Language: Python
File: inference/worker/utils.py
PR Title: Cleaned up backend plugin code: Part 1
BEFORE CODE: ids = ids[-(max_input_length - 1) :]
with shared_tokenizer_lock:
prompt = tokenizer.decode(ids)
prompt = add_prefix_to_prompt(prompt... | I think this one should remain `warning`, it's relevant outside of debugging. | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
import { Badge } from "@chakra-ui/react";
// kept brief so that it doesnt distract/interfere with user experience
export const EncourageMessage = () => {
return (
<Badge ml="1" fontFamily="Geneva" fontWeight="normal" fontSize="0.4em" background="gray.300" color="black">
Thoughts? Let us know!
</Badge>
... | can you move this to the localization? | import { Badge } from "@chakra-ui/react";
import { useTranslation } from "next-i18next";
// kept brief so that it doesnt distract/interfere with user experience
export const EncourageMessage = () => {
const { t } = useTranslation("chat");
return (
<Badge fontWeight="normal" background="gray.300" color="black">
... | @@ -0,0 +1,10 @@
+import { Badge } from "@chakra-ui/react";
+
+// kept brief so that it doesnt distract/interfere with user experience
+export const EncourageMessage = () => {
+ return (
+ <Badge ml="1" fontFamily="Geneva" fontWeight="normal" fontSize="0.4em" background="gray.300" color="black">
+ Thoughts? Le... | website/src/components/Chat/EncourageMessage.tsx | 7 | TypeScript | 0.143 | refactor | 38 | 11 | 12 | false | Created feedback encouragement message | 3,057 | LAION-AI/Open-Assistant | 29,137 | Python | AbdBarho | amitag2 | Review this pull request change.
Language: TypeScript
File: website/src/components/Chat/EncourageMessage.tsx
PR Title: Created feedback encouragement message
BEFORE CODE: import { Badge } from "@chakra-ui/react";
// kept brief so that it doesnt distract/interfere with user experience
export const EncourageMessage ... | can you move this to the localization? | You are an expert software engineer and code reviewer. Review pull requests carefully for correctness, bugs, security issues, performance problems, maintainability, and unnecessary complexity. Only raise actionable issues supported by the code. Explain why an issue matters and suggest a concrete fix when appropriate. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.