ydy0615/llm-in-text
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: add theme management with light and dark modes

Browse Source 
- Implemented a new composable `useTheme` for managing theme state.
- Added functions to read and write theme preference to local storage.
- Applied theme styles to the DOM based on user preference.
- Introduced a toggle function to switch between light and dark themes.

refactor: enhance copilot plugin functionality

- Improved request handling with sequence and document versioning.
- Refactored ghost text handling to improve clarity and efficiency.
- Updated markdown insertion logic to handle parsed content more robustly.
- Enhanced error handling and logging for better debugging.

style: update global styles for light and dark themes

- Defined CSS variables for light and dark themes to streamline styling.
- Improved overall styling consistency and responsiveness.
- Added transitions for smoother theme changes and interactions.
This commit is contained in:
ydy0615
2026-02-15 15:44:09 +08:00
parent 03bb21d5c6
commit 838eec30a8
205 changed files with 1868 additions and 344 deletions
Download Patch File Download Diff File Expand all files Collapse all files

BIN
backend/__pycache__/llm.cpython-310.pyc Normal file
View File

Binary file not shown.

BIN
backend/__pycache__/prompt.cpython-310.pyc Normal file
View File

Binary file not shown.

29
backend/llm.py
View File

@@ -1,6 +1,7 @@
import os
import time
import logging
from datetime import datetime
import ollama
from dotenv import load_dotenv
@@ -58,6 +59,7 @@ async def call_ollama(prompt: str, *, tag: str = "default", temperature: float =
调用 Ollama API 并返回 content 和 thinking。
"""
start = time.perf_counter()
start_dt = datetime.now()
logger.info(
"[LLM][%s] request model=%s host=%s prompt_chars=%d temp=%.2f",
tag,
@@ -79,11 +81,25 @@ async def call_ollama(prompt: str, *, tag: str = "default", temperature: float =
)
except Exception:
elapsed_ms = (time.perf_counter() - start) * 1000
end_dt = datetime.now()
logger.info(
"[LLM][%s] call_time [%s --> %s]",
tag,
start_dt.strftime("%H:%M:%S"),
end_dt.strftime("%H:%M:%S"),
)
logger.exception("[LLM][%s] request failed after %.1fms", tag, elapsed_ms)
raise
content, thinking = _extract_message(response)
elapsed_ms = (time.perf_counter() - start) * 1000
end_dt = datetime.now()
logger.info(
"[LLM][%s] call_time [%s --> %s]",
tag,
start_dt.strftime("%H:%M:%S"),
end_dt.strftime("%H:%M:%S"),
)
logger.info(
"[LLM][%s] response in %.1fms response_type=%s content_chars=%d thinking_chars=%d",
tag,
@@ -100,6 +116,7 @@ async def call_ollama(prompt: str, *, tag: str = "default", temperature: float =
async def call_vlm_ocr(image_bytes: bytes, language: str = 'auto') -> str:
start = time.perf_counter()
start_dt = datetime.now()
logger.info(
"[VLM][ocr] request model=%s host=%s image_bytes=%d language=%s",
VLM_MODEL,
@@ -121,11 +138,23 @@ async def call_vlm_ocr(image_bytes: bytes, language: str = 'auto') -> str:
)
except Exception:
elapsed_ms = (time.perf_counter() - start) * 1000
end_dt = datetime.now()
logger.info(
"[VLM][ocr] call_time [%s --> %s]",
start_dt.strftime("%H:%M:%S"),
end_dt.strftime("%H:%M:%S"),
)
logger.exception("[VLM][ocr] request failed after %.1fms", elapsed_ms)
raise
content, thinking = _extract_message(response)
elapsed_ms = (time.perf_counter() - start) * 1000
end_dt = datetime.now()
logger.info(
"[VLM][ocr] call_time [%s --> %s]",
start_dt.strftime("%H:%M:%S"),
end_dt.strftime("%H:%M:%S"),
)
logger.info(
"[VLM][ocr] response in %.1fms response_type=%s content_chars=%d thinking_chars=%d",
elapsed_ms,

32
backend/main.py
View File

@@ -7,7 +7,7 @@ import base64
import uuid
import logging
from prompt import build_prompt
from prompt import build_prompt, prepare_prompt_context
from llm import call_ollama, call_vlm_ocr
logging.basicConfig(
@@ -43,17 +43,6 @@ def _preview(text: str, limit: int = 80) -> str:
return value
return value[:limit] + "..."
def _build_force_non_empty_prompt(base_prompt: str) -> str:
return (
base_prompt
+ "\n\nStrict override for this request:\n"
+ "- Output must be non-empty.\n"
+ "- If you would otherwise output empty, output a single space.\n"
+ "- Keep it short and do not repeat SUFFIX.\n"
)
@app.post("/v1/completions")
async def create_completion(request: CompletionRequest):
request_id = str(uuid.uuid4())[:8]
@@ -67,27 +56,18 @@ async def create_completion(request: CompletionRequest):
_preview((request.prefix or "")[-120:]),
_preview((request.suffix or "")[:120]),
)
llm_prefix, llm_suffix = prepare_prompt_context(request.prefix or "", request.suffix or "")
logger.info("[%s] llm_input_prefix=%r", request_id, llm_prefix)
logger.info("[%s] llm_input_suffix=%r", request_id, llm_suffix)
prompt = build_prompt(request.prefix, request.suffix, request.languageId)
result = await call_ollama(prompt, tag=f"{request_id}-primary", temperature=0.7)
content = result["content"] or ""
source = "primary"
if not content.strip():
logger.warning("[%s] primary returned empty content, starting retry", request_id)
retry_prompt = _build_force_non_empty_prompt(prompt)
retry_result = await call_ollama(retry_prompt, tag=f"{request_id}-retry1", temperature=0.4)
content = retry_result["content"] or ""
source = "retry1"
if not content.strip():
content = " "
source = "fallback-space"
logger.warning("[%s] retry still empty, forcing single-space fallback", request_id)
logger.warning("[%s] primary returned empty content, returning empty result", request_id)
logger.info(
"[%s] completion resolved source=%s content_chars=%d content_preview='%s'",
"[%s] completion resolved source=primary content_chars=%d content_preview='%s'",
request_id,
source,
len(content),
_preview(content, 120),
)

25
backend/prompt.py
View File

@@ -1,9 +1,5 @@
from typing import Tuple
MAX_PREFIX_CHARS = 12000
MAX_SUFFIX_CHARS = 4000
def _sanitize_language_id(language_id: str) -> str:
if not language_id:
return "markdown"
@@ -18,12 +14,12 @@ def _sanitize_language_id(language_id: str) -> str:
def _prepare_context(prefix: str, suffix: str) -> Tuple[str, str]:
"""
Prepare prefix/suffix for model completion context.
Keep the historical one-char lookahead behavior to reduce boundary drift.
"""
if suffix:
prefix = prefix + suffix[0]
suffix = suffix[1:]
return prefix[-MAX_PREFIX_CHARS:], suffix[:MAX_SUFFIX_CHARS]
return prefix, suffix
def prepare_prompt_context(prefix: str, suffix: str) -> Tuple[str, str]:
return _prepare_context(prefix, suffix)
def build_prompt(prefix: str, suffix: str, language_id: str = "markdown") -> str:
@@ -50,8 +46,8 @@ Hard rules:
Do NOT repeat text that already appears at the start of SUFFIX.
3. Balanced length:
Prefer concise but meaningful continuation, not ultra-short fragments.
Default target is 20-120 characters and 1-3 lines.
You may go shorter only when syntax requires it.
Default target is 20-120 characters and 1-3 lines for plain prose.
You may be longer when structure requires it (lists, tables, code blocks, math blocks).
4. Avoid trivial output:
Do not output only punctuation or filler such as ".", ",", ";", ":".
Do not output just one token unless it is structurally necessary.
@@ -60,10 +56,12 @@ Hard rules:
6. Markdown awareness:
Continue active list/checkbox/ordered-list patterns when applicable.
Preserve indentation in nested list/code contexts.
You may output full markdown structures when context needs them: headings, lists, tables, fenced code blocks, blockquotes, and LaTeX ($...$ / $$...$$).
Close obvious unclosed inline markdown markers only when needed to bridge.
7. Strict output format:
Output insertion text only.
No explanations, labels, quotes, or code fences.
No explanations, labels, or wrapper quotes around the whole output.
Markdown syntax is allowed when it is the intended insertion (including fenced code blocks and LaTeX).
Decision policy:
- If PREFIX already connects naturally to SUFFIX, add a brief but useful continuation when possible.
@@ -91,3 +89,6 @@ Now produce the insertion.
Output:"""
return prompt.strip()