backend/__pycache__/llm.cpython-313.pyc

<EFBFBD>
<00>?<3F>iz<00>	<00>N<00>SSKrSSKrSSKrSSKrSSKJr \"5 \R"SS5r\R"SS5r\R"SS5r	\R"\S	9r\R"S
5r
SrS\\\44S
jrSSS.S\S\S\S\4SjjrSS\S\S\4Sjjrg)<19>N)<01>load_dotenv<6E>OLLAMA_MODELzgpt-oss:20b<30>OLLAMA_HOSTzhttp://192.168.0.120:11434<33>	VLM_MODELzqwen3-vl:30b)<01>host<73>llmaYou are an OCR and visual-context extractor for markdown writing assistance.

Your output will be embedded inside an HTML comment as hidden context for a text-completion model.

Requirements:
- Keep output compact: maximum 120 words.
- Use plain text only (no markdown code fences).
- Never output <!-- or -->.
- Do not invent unreadable text; mark uncertain characters with ?.
- Preserve original script for recognized text (do not forcibly translate).

Return exactly this format:

TEXT:
<exact transcription of visible text; use " | " for line breaks; write "(none)" if no readable text>

KEY_DETAILS:
- <3-5 short factual bullets about relevant objects/layout>

LANGUAGE:
<dominant language(s) in visible text, e.g. English / Chinese / Mixed>

SUMMARY:
<one short sentence, <= 20 words><3E>returnc<00><><00>SnSn[US5(aSUR(aBURR=(d Sn[URSS5=(d SnX4$[	U[
5(aHUR
S05nUR
SS5=(d SnUR
SS5=(d SnX4$)N<><00>message<67>thinking<6E>content)<07>hasattrrr<00>getattr<74>
isinstance<EFBFBD>dict<63>get)<04>responserr
<00>msgs    <20>+/Users/allenyuan/llm-in-text/backend/llm.py<70>_extract_messager)s<><00><00><10>G<EFBFBD><11>H<EFBFBD><0E>x<EFBFBD><19>#<23>#<23><08>(8<>(8<><1A>"<22>"<22>*<2A>*<2A>0<>b<EFBFBD><07><1A>8<EFBFBD>+<2B>+<2B>Z<EFBFBD><12><<3C>B<><02><08><13><1C><1C>
<14>H<EFBFBD>d<EFBFBD>	#<23>	#<23><16>l<EFBFBD>l<EFBFBD>9<EFBFBD>b<EFBFBD>)<29><03><15>'<27>'<27>)<29>R<EFBFBD>(<28>.<2E>B<EFBFBD><07><16>7<EFBFBD>7<EFBFBD>:<3A>r<EFBFBD>*<2A>0<>b<EFBFBD><08><12><1C><1C><00>defaultgffffff<66>?)<02>tag<61>temperature<72>promptrrc
<00><00><># <00>[R"5n[RSU[[
[
U5U5 [R[SUS./SUSS.S9IShv<01>Nn[U5upg[R"5U-
S	-n[RSUU[U5R[
U5[
U55 UR5(d[RSU5 XgS
.$N<>![a3 [R"5U-
S	-n[RS
X5 ef=f7f)u5
调用 Ollama API 并返回 content 和 thinking。
z<[LLM][%s] request model=%s host=%s prompt_chars=%d temp=%.2f<EFBFBD>user)<02>rolerFg<46><67><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>?)r<00>repeat_penalty<74><04>model<65>messages<65>stream<61>optionsN<73><4E>z%[LLM][%s] request failed after %.1fmszP[LLM][%s] response in %.1fms response_type=%s content_chars=%d thinking_chars=%dz)[LLM][%s] empty content returned by model)rr
)<10>time<6D>perf_counter<65>logger<65>inforr<00>len<65>client<6E>chat<61>	Exception<6F>	exceptionr<00>type<70>__name__<5F>strip<69>warning)rrr<00>startr<00>
elapsed_msrr
s        r<00>call_ollamar68s-<00><00><00>
<11><1D><1D><1F>E<EFBFBD>
<EFBFBD>K<EFBFBD>K<EFBFBD>F<><0B><14><13><0B>F<EFBFBD><0B><13>
<06>
<0E><1F><1B><1B><1E>%<25>&<26>9<>:<3A><18>*<2A>"%<25><0E>	%<25>
<EFBFBD>
<EFBFBD><08>)<29><18>2<><15>G<EFBFBD><16>#<23>#<23>%<25><05>-<2D><14>5<>J<EFBFBD>
<EFBFBD>K<EFBFBD>K<EFBFBD>Z<><0B><12><0C>X<EFBFBD><0E><1F><1F><0B>G<EFBFBD><0C><0B>H<EFBFBD>
<0A>
<06><13>=<3D>=<3D>?<3F>?<3F><0E><0E><0E>B<>C<EFBFBD>H<><1E>5<>5<>9
<EFBFBD><EFBFBD><15><0E><1A>'<27>'<27>)<29>E<EFBFBD>1<>T<EFBFBD>9<>
<EFBFBD><0E><18><18>@<40>#<23>R<>
<0A><0E>s1<00>AE<01>$D<00>(D<04>)D<00>-BE<01>D<00>=E<03>E<01>image_bytes<65>languagec	<00><00><># <00>[R"5n[RS[[
[
U5U5 [R[S[U/S./SSS0S9IShv<01>Nn[U5upV[R"5U-
S-n[RS
U[U5R[
U5[
U55 UR5(d[R!S5 U$N<>![a3 [R"5U-
S-n[RS	U5 ef=f7f)Nz>[VLM][ocr] request model=%s host=%s image_bytes=%d language=%sr)rr<00>imagesFrg333333<33>?r!r&z&[VLM][ocr] request failed after %.1fmszQ[VLM][ocr] response in %.1fms response_type=%s content_chars=%d thinking_chars=%dz*[VLM][ocr] empty content returned by model)r'r(r)r*rrr+r,r-<00>VLM_OCR_CONTEXT_PROMPTr.r/rr0r1r2r3)r7r8r4rr5rr
s       r<00>call_vlm_ocrr<es)<00><00><00><10><1D><1D><1F>E<EFBFBD>
<EFBFBD>K<EFBFBD>K<EFBFBD>H<><11><13><0B>K<EFBFBD><18><10><06><0E><1F><1B><1B><1B><1E>1<>&<26>-<2D><0E><0F>
<19>"<22>C<EFBFBD>(<28>%<25>	
<EFBFBD>	
<EFBFBD><08>)<29><18>2<><15>G<EFBFBD><16>#<23>#<23>%<25><05>-<2D><14>5<>J<EFBFBD>
<EFBFBD>K<EFBFBD>K<EFBFBD>[<5B><12><0C>X<EFBFBD><0E><1F><1F><0B>G<EFBFBD><0C><0B>H<EFBFBD>
<0A><06><13>=<3D>=<3D>?<3F>?<3F><0E><0E><0E>C<>D<><12>N<EFBFBD>9	
<EFBFBD><EFBFBD><15><0E><1A>'<27>'<27>)<29>E<EFBFBD>1<>T<EFBFBD>9<>
<EFBFBD><0E><18><18>A<>:<3A>N<>
<0A><0E>s1<00>AE<01>)D<00>,D<04>-D<00>1BE<01>D<00>=E<03>E)<01>auto)<17>osr'<00>logging<6E>ollama<6D>dotenvr<00>getenvrrr<00>AsyncClientr,<00>	getLoggerr)r;<00>tuple<6C>strr<00>floatrr6<00>bytesr<<00>rr<00><module>rJs<><00><01>	<09><0B><0E>
<0A><1E><0B>
<0A><11>y<EFBFBD>y<EFBFBD><1E><1D>7<><0C><10>i<EFBFBD>i<EFBFBD>
<0A>'C<>D<><0B><0E>I<EFBFBD>I<EFBFBD>k<EFBFBD>><3E>2<>	<09>	<0F>	<1B>	<1B><1B>	-<2D><06>	<10>	<1A>	<1A>5<EFBFBD>	!<21><06>%<25><16>2<1D>%<25><03>S<EFBFBD><08>/<2F><1D>2;<3B>QT<51>+6<>c<EFBFBD>+6<>3<EFBFBD>+6<><15>+6<>Y]<5D>+6<>Z'<13>E<EFBFBD>'<13>S<EFBFBD>'<13>c<EFBFBD>'r
-												feat: switch from OpenAI API to native Ollama Python client

This commit refactors the LLM integration to use Ollama's native Python client instead of OpenAI-compatible API, while fixing critical template syntax errors and improving project structure.

Key changes:
- Replace openai package with ollama package in backend requirements
- Rewrite llm.py to use ollama.AsyncClient for direct Ollama API calls
- Update main.py to use non-streaming Ollama responses with thinking extraction
- Fix template syntax error in MilkdownEditor.vue (GhostTextOverlay component tags)
- Fix string截取错误 by using slice() instead of substring()
- Add src/utils/api.js and src/utils/config.js for shared configuration
- Add CORS middleware to FastAPI backend
- Update prompt.py with clearer instructions for continuation generation
- Add comprehensive README.md documentation

BREAKING CHANGE: Environment variables OLLAMA_BASE_URL changed to OLLAMA_HOST (remove /v1/ suffix)

											
										
										
											2026-02-07 08:53:37 +08:00
+								<EFBFBD>
-												feat(editor): add image insertion with OCR support and size limit handling

Add image button with dropdown menu for uploading local images or inserting from URL.
Integrate VLM-based OCR to extract text context from images and include in AI suggestions.
Implement document size limits to disable AI when exceeding threshold.
Refactor copilot plugin with per-view runtime state and OCR context injection.
Add OCR cache utility for managing image metadata.
Add code splitting configuration for optimized bundle size.

											
										
										
											2026-02-14 18:28:37 +08:00
+								<00>?<3F>iz<00>	<00>N<00>SSKrSSKrSSKrSSKrSSKJr \"5 \R"SS5r\R"SS5r\R"SS5r	\R"\S	9r\R"S
 r
SrS\\\44S
jrSSS.S\S\S\S\4SjjrSS\S\S\4Sjjrg)<19>N)<01>load_dotenv<6E>OLLAMA_MODELzgpt-oss:20b<30>OLLAMA_HOSTzhttp://192.168.0.120:11434<33>	VLM_MODELzqwen3-vl:30b)<01>host<73>llmaYou are an OCR and visual-context extractor for markdown writing assistance.
 								Your output will be embedded inside an HTML comment as hidden context for a text-completion model.
 								Requirements:
 								- Keep output compact: maximum 120 words.
 								- Use plain text only (no markdown code fences).
 								- Never output <!-- or -->.
 								- Do not invent unreadable text; mark uncertain characters with ?.
 								- Preserve original script for recognized text (do not forcibly translate).
 								Return exactly this format:
 								TEXT:
 								<exact transcription of visible text; use " | " for line breaks; write "(none)" if no readable text>
 								KEY_DETAILS:
 								- <3-5 short factual bullets about relevant objects/layout>
 								LANGUAGE:
 								<dominant language(s) in visible text, e.g. English / Chinese / Mixed>
 								SUMMARY:
 								<one short sentence, <= 20 words><3E>returnc<00><><00>SnSn[US5(aSUR(aBURR=(d Sn[URSS5=(d SnX4$[	U[
 (aHUR
S05nUR
SS5=(d SnUR
SS5=(d SnX4$)N<><00>message<67>thinking<6E>content)<07>hasattrrr<00>getattr<74>
 								isinstance<EFBFBD>dict<63>get)<04>responserr
<00>msgs    <20>+/Users/allenyuan/llm-in-text/backend/llm.py<70>_extract_messager)s<><00><00><10>G<EFBFBD><11>H<EFBFBD><0E>x<EFBFBD><19>#<23>#<23><08>(8<>(8<><1A>"<22>"<22>*<2A>*<2A>0<>b<EFBFBD><07><1A>8<EFBFBD>+<2B>+<2B>Z<EFBFBD><12><<3C>B<><02><08><13><1C><1C>
 								<14>H<EFBFBD>d<EFBFBD>	#<23>	#<23><16>l<EFBFBD>l<EFBFBD>9<EFBFBD>b<EFBFBD>)<29><03><15>'<27>'<27>)<29>R<EFBFBD>(<28>.<2E>B<EFBFBD><07><16>7<EFBFBD>7<EFBFBD>:<3A>r<EFBFBD>*<2A>0<>b<EFBFBD><08><12><1C><1C><00>defaultgffffff<66>?)<02>tag<61>temperature<72>promptrrc
 								<00><00><># <00>[R"5n[RSU[[
 								[
U5U5 [R[SUS./SUSS.S9IShv<01>Nn[U5upg[R"5U-
 								S	-n[RSUU[U5R[
U5[
U55 UR5(d[RSU5 XgS
.$N<>![a3 [R"5U-
 								S	-n[RS
 								X5 ef=f7f)u5
 								调用 Ollama API 并返回 content 和 thinking。
 								z<[LLM][%s] request model=%s host=%s prompt_chars=%d temp=%.2f<EFBFBD>user)<02>rolerFg<46><67><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>?)r<00>repeat_penalty<74><04>model<65>messages<65>stream<61>optionsN<73><4E>z%[LLM][%s] request failed after %.1fmszP[LLM][%s] response in %.1fms response_type=%s content_chars=%d thinking_chars=%dz)[LLM][%s] empty content returned by model)rr
)<10>time<6D>perf_counter<65>logger<65>inforr<00>len<65>client<6E>chat<61>	Exception<6F>	exceptionr<00>type<70>__name__<5F>strip<69>warning)rrr<00>startr<00>
 								elapsed_msrr
s        r<00>call_ollamar68s-<00><00><00>
<11><1D><1D><1F>E<EFBFBD>
 								<EFBFBD>K<EFBFBD>K<EFBFBD>F<><0B><14><13><0B>F<EFBFBD><0B><13>
<06>
<0E><1F><1B><1B><1E>%<25>&<26>9<>:<3A><18>*<2A>"%<25><0E>	%<25>
 								<EFBFBD>
 								<EFBFBD><08>)<29><18>2<><15>G<EFBFBD><16>#<23>#<23>%<25><05>-<2D><14>5<>J<EFBFBD>
 								<EFBFBD>K<EFBFBD>K<EFBFBD>Z<><0B><12><0C>X<EFBFBD><0E><1F><1F><0B>G<EFBFBD><0C><0B>H<EFBFBD>
<0A>
<06><13>=<3D>=<3D>?<3F>?<3F><0E><0E><0E>B<>C<EFBFBD>H<><1E>5<>5<>9
 								<EFBFBD><EFBFBD><15><0E><1A>'<27>'<27>)<29>E<EFBFBD>1<>T<EFBFBD>9<>
 								<EFBFBD><0E><18><18>@<40>#<23>R<>
<0A><0E>s1<00>AE<01>$D<00>(D<04>)D<00>-BE<01>D<00>=E<03>E<01>image_bytes<65>languagec	<00><00><># <00>[R"5n[RS[[
 								[
U5U5 [R[S[U/S./SSS0S9IShv<01>Nn[U5upV[R"5U-
 								S-n[RS
 								U[U5R[
U5[
U55 UR5(d[R!S5 U$N<>![a3 [R"5U-
 								S-n[RS	U5 ef=f7f)Nz>[VLM][ocr] request model=%s host=%s image_bytes=%d language=%sr)rr<00>imagesFrg333333<33>?r!r&z&[VLM][ocr] request failed after %.1fmszQ[VLM][ocr] response in %.1fms response_type=%s content_chars=%d thinking_chars=%dz*[VLM][ocr] empty content returned by model)r'r(r)r*rrr+r,r-<00>VLM_OCR_CONTEXT_PROMPTr.r/rr0r1r2r3)r7r8r4rr5rr
s       r<00>call_vlm_ocrr<es)<00><00><00><10><1D><1D><1F>E<EFBFBD>
 								<EFBFBD>K<EFBFBD>K<EFBFBD>H<><11><13><0B>K<EFBFBD><18><10><06><0E><1F><1B><1B><1B><1E>1<>&<26>-<2D><0E><0F>
 								<19>"<22>C<EFBFBD>(<28>%<25>
 								<EFBFBD>
 								<EFBFBD><08>)<29><18>2<><15>G<EFBFBD><16>#<23>#<23>%<25><05>-<2D><14>5<>J<EFBFBD>
 								<EFBFBD>K<EFBFBD>K<EFBFBD>[<5B><12><0C>X<EFBFBD><0E><1F><1F><0B>G<EFBFBD><0C><0B>H<EFBFBD>
<0A><06><13>=<3D>=<3D>?<3F>?<3F><0E><0E><0E>C<>D<><12>N<EFBFBD>9
 								<EFBFBD><EFBFBD><15><0E><1A>'<27>'<27>)<29>E<EFBFBD>1<>T<EFBFBD>9<>
 								<EFBFBD><0E><18><18>A<>:<3A>N<>
<0A><0E>s1<00>AE<01>)D<00>,D<04>-D<00>1BE<01>D<00>=E<03>E)<01>auto)<17>osr'<00>logging<6E>ollama<6D>dotenvr<00>getenvrrr<00>AsyncClientr,<00>	getLoggerr)r;<00>tuple<6C>strr<00>floatrr6<00>bytesr<<00>rr<00><module>rJs<><00><01>	<09><0B><0E>
<0A><1E><0B>
<0A><11>y<EFBFBD>y<EFBFBD><1E><1D>7<><0C><10>i<EFBFBD>i<EFBFBD>
<0A>'C<>D<><0B><0E>I<EFBFBD>I<EFBFBD>k<EFBFBD>><3E>2<>	<09>	<0F>	<1B>	<1B><1B>	-<2D><06>	<10>	<1A>	<1A>5<EFBFBD>	!<21><06>%<25><16>2<1D>%<25><03>S<EFBFBD><08>/<2F><1D>2;<3B>QT<51>+6<>c<EFBFBD>+6<>3<EFBFBD>+6<><15>+6<>Y]<5D>+6<>Z'<13>E<EFBFBD>'<13>S<EFBFBD>'<13>c<EFBFBD>'r