LM Studioã§åœç£OSSå€§èŠæš¡èšèªã¢ãã«LLM-jp4ã䜿ã
æè¿ããã¥ãŒã¹ãèŠãŠãããšãåœç£LLMãåœç«æ å ±åŠç ç©¶æ(NII)ãäœããLLM-jpãšããåã§OSSã«ããŠããããšãç¥ããŸããã
ããããã®ã欲ããã£ãããšããããšã§ãããŒã«ã«LLMã䜿ããããããŠããããœãã LM Studio ããæ©é䜿ã£ãŠã¿ãŸãããããã®å€ãããœã³ã³ã§ãåãã®ãã»ã»ã»ïŒåããŸããïŒã
ã¹ããã¯
Windows11 Pro
GPU: Nvidia GeForce RTX 3060 12G
ã¡ã¢ãª: 64G
LM Studio 0.4.12
LLM-jpãšã¯ïŒ
æ¥æ¬çºã®LLMéçºãããžã§ã¯ããç ç©¶è ã³ãã¥ããã£åã
2026幎4æ3æ¥ã« llm-jp-4 ã·ãªãŒãº 8b, 32b ãå ¬éããããã¢ãŒããã¯ãã£ã¯llama2, qwen3moeã䜿ã£ãŠããããåŠç¿èªäœã¯ãã«ã¹ã¯ã©ãããGPT-4oãäžåãæ§èœãæ¥æ¬èªã§èŠãããåçšå©çšãå¯èœãªOSS(Apache2.0)ã
LLM-jp-4 玹ä»ããæç² https://drive.google.com/file/d/1G6CrlVHa73zuJPxgRbqbiec1tRU7_wGj/view
ã¡ãªã¿ã«ã次ã®ããŒãžã«å匷äŒã®è³æãå ¬éãããŠããã®ã§ãããªãå匷ã«ãªãïŒã¡ãã£ãšèªãã ã ãã§ããïŒã
LM Studioãšã¯
ããŒã«ã«ã¢ãã«ãç°¡åã«è©Šãããœããã§ããç»é¢äžããç°¡åã«ããŒã«ã«ã¢ãã«ãããŠã³ããŒãããŠããã£ããç»é¢ã§äœ¿ããŸãããŸããããŒã«ã«APIïŒãšãã£ãŠããã®ãïŒïŒãšããŠäœ¿ããã²ãšãŸãã¢ãã«ãããŠã³ããŒãããŠAPIãã䜿ãã¿ãããªããšãã§ããŸãã
ãã£ããç»é¢

ã¡ãªã¿ã«ãLM Studioã«äŒŒããœãããšããŠã¯å¥ã« ollama ããããã¡ãªã¿ã«çè ã¯ä»¥åãollamaã䜿ã£ãŠããããä»åã¯ã¡ãã£ãšæ°å転æãšããããšã§ LM Studio ã䜿ã£ãŠã¿ãããã©ã¡ããã©ãããšããã»ã©äœ¿ã蟌ãã§ãªãã®ã§ãã©ã¡ãããããšæãã
ã¡ãªã¿ã«ããã«é¢ããå·ç(å®å šã«å¿ããŠãã):
LLM-jp4ãããŠã³ããŒã
ããŠãä»åã¯ãšãããã䜿ã£ãŠã¿ãŸãããŸãã¯ã¢ãã«ã®ããŠã³ããŒãã§ãã
å·ŠåŽã®ã¡ãã¥ãŒã®ããããæ€çŽ¢ã®ããŒã¯ïŒäžããïŒã€ç®ïŒãã¯ãªãã¯ããã¢ãã«ãæ€çŽ¢ããŸããããããåºãŠããŸãã

ã¢ãã«åã®åŸãã«ggufãšã€ããŠããã®ã§ãããããã¯LM Studioãªã©ã§æ±ããããã«å€æããã圢åŒã ããã§ãïŒby gemini)ãhuggingfaceã«å ¬åŒã®ã¢ã«ãŠã³ããããã®ã«ãéãã¢ã«ãŠã³ãåã§åºãŠããã®ã¯ãªãã ããšæã£ãã®ã§ããã䜿ãã圢ã«å€æããŠãã ãã£ãŠããšããããšã§ããããããããŸã§ç°¡åã«äœ¿ããŸããããããšãããããŸãã


ããŠã³ããŒãã®ãšãããèŠããšã32bã¯21GBã8bã¯5GBãšãªããŸãã倧ãããªããã§ã䜿ãããã®ã§ããŠã³ããŒãã
LLM-jp4ã䜿ã
ããŠã³ããŒããããŒãã䜿ããïŒïŒïŒãšããæãã§ãããæ®éã«2åç®ã®ãã£ãããããšãšã©ãŒãåºãŠããŸããã

Error rendering prompt with jinja template: "You have passed a message containing <|channel|> tags in the content field. Instead of doing this, you should pass analysis messages (the string between '<|message|>' and '<|end|>') in the 'thinking' field, and final messages (the string between '<|message|>' and '<|end|>') in the 'content' field.".
This is usually an issue with the model's prompt template. If you are using a popular model, you can try to search the model under lmstudio-community, which will have fixed prompt templates. If you cannot find one, you are welcome to post this issue to our discord or issue tracker on GitHub. Alternatively, if you know how to write jinja templates, you can override the prompt template in My Models > model settings > Prompt Template.
ã²ããŒãŒãŒãã©ããã£ãŠåãããã»ã»ã»ããã³ãã¬ãŒãã¯è§Šããããªããªããšããããšã§ãããã¯æçµææ®µãšããããšãšããŸããã<|channel|>ã£ãŠå ¥ã£ãŠãªãã£ããåãã®ããªãã»ã»ã»
解決çãæ¢ã
ãšããããšã§ãåšèŸºç¥èã調ã¹ãŠãããšã8Bã«æ¬¡ã®ãããªã³ã¡ã³ãããããŸããã
LMStudio
<|channel|>final<|message|>ãLMStudioäžã§åºãŠããŸããŸãã®ã§ãMyModels>ã¢ãã«>Inference>Reasoning Parsingããã
éå§æåå <|channel|> çµäºæåå final<|message|> ãèšå®ãããšæ¶ããŸãããææ°ãããããŸã.
ãªãã»ã©ããšããããšã§ã8BãåãããŠã¿ããšããªããåããã»ã»ã»ãã§ããstart stringãæ¬¡ã®ããã«ãããåãã ããŸããã<|channel>ãå¶ç¶ãªã£ãŠããã®ããçã£ãŠãªã£ãã®ãåãããªããã©ãæ°ä»ããæã«ã¯ããã§åããŠããŸããã32Bãããã§åããŸããã
ãã®èŸºã¯ãã£ãããŒãšè©±ããŠãããšãreasoningãäŒè©±ã«å ¥ãããã¿ãããªãšããã§ãLM Studioã¯Jinja templateã« <|Channel|>ãšå ¥ã£ãŠãããšã©ãŒã«ããã¿ãããªè©±é¡ãšãªããŸããã
LLM-jp4ãåãã
ãšããèš³ã§åãããŠã¿ãŸããã
8B
ãŸãã¯8bãããããŒããªãšãã£ãåŸã次ã®èª²é¡ãæž¡ããŠã¿ãŸããã
ããååºäº€æã®ã瀌ã以äžã®æ¡ä»¶ãæºããã¡ãŒã«ãšããŠäœæããŠãã ããã
200æå以å
æ¬¡åæ¥çšåè£ã2ã€å
¥ãã (2026幎5æã§)
âæãå
¥ããŸããâã1å䜿ã
ç®æ¡æžãçŠæ¢
ã
ãããšæ¬¡ã®ãããªè¿ä¿¡ã§ããã
ïŒã€ç®: ãããã«ã¡ã¯ïŒ
äœãã質åãçžè«ãããããšãããã°ãé æ
®ãªãæããŠãã ããããðã
ïŒã€ç®: ãããæ§
æ¬æ¥ã¯ãå¿ãããšããååºäº€æã®æ©äŒãããã ããèª ã«ããããšãããããŸãããæãå
¥ããŸããæ¬¡åã®æã¡åããã«ã€ããŸããŠã以äžã®æ¥çšããæç€ºããããŸãã®ã§ã確èªãã ãããâ 2026幎5æ10æ¥ïŒç«ïŒåå10æãâ¡2026幎5æ18æ¥ïŒæ°ŽïŒååŸ2æããéœåã®è¯ãæ¥æããè¿ä¿¡ããã ããŸããšå¹žãã§ãããããããé¡ãããŸãã
éåžžã«éãã¹ããŒãã§ã20ç§ãããã§çµäº
ã¹ããŒã確èªåç»:
32B
32Bã«ãåãå å®¹ãæž¡ãããšããããªããšãïŒã€ç®ã®è³ªåãæ°žé ãšèãç¶ããŠããŸããŸãããã©ãããããã»ã»ã»ãããããããreasoning parsing ãããŸãè¡ã£ãŠãªããããããããŸãããã»ã»ã»ã
ïŒã€ç®: ã ããã«ã¡ã¯ïŒãããŒããªãã§ããã仿¥ã¯ã©ã®ãããªãæäŒããããŸããããïŒã
ïŒã€ç®: ãããæ ªåŒäŒç€Ÿãâ³â³æ§
ãã€ããäžè©±ã«ãªã£ãŠãããŸããå
æ¥ã¯ååºãã亀æããã ããèª ã«ããããšãããããŸããã
æãå
¥ããŸããæ¬¡åã®æã¡åããã«ã€ããŸããŠã5æ10æ¥ïŒç«ïŒãŸãã¯5æ17æ¥ïŒæ°ŽïŒã®ãããããéœåã¯ãããã§ããããããè¿ä¿¡ãåŸ
ã¡ããŠãããŸãã
å±±ç°å€ªé
æ ªåŒäŒç€ŸABCã
ãããªé
ãã¯ãªãããæèãè²ã
ãªãšããã«é£ã³ãïŒåãããã§çµäºãæåãæ°ãããããããããŠããã
ã¹ããŒã確èªåç»:
ãŸãšã
åLLM-jp4ã䜿ã£ãŠã¿ãŸãããããŒã«ã«LLMã ãã§ã§ãããã®ãèŠã€ãããããšé¢çœãããã«ãæ¥æ¬èªã«åŒ·ããšããã®ã¯éåžžã«é¢çœããããã¯ãã«ãªãå¯èœæ§ããããOSSãšããããšã§ãæŽ»çšæ¹æ³ãããããèããŠããŸãã
æè¿ OpenAI ã privacy filter modelãšããã®ãåºããŠãŸããããããã¡ãã£ãšäœ¿ã£ããšãããæ¥æ¬èªã¯ãŸã ããŸãã§ããªãæãã ã£ãã®ã§ãããããçšéã§ãé¢çœãããªãšæã£ããããŠããŸãã
Discussion