[B! GPU] deejayrokaã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

deejayroka id:deejayroka

GPUã«é–¢ã™ã‚‹deejayrokaã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (17)

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

ãƒ–ãƒ©ã‚¦ã‚¶ã§LLMã‚’ãƒãƒ¼ã‚«ãƒ«å±•é–‹ã™ã‚‹æ‰‹æ³• | AIDB
æœ¬è¨˜äº‹ã§ã¯ã€ãƒ–ãƒ©ã‚¦ã‚¶å†…ã§ã®å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ï¼ˆLLMï¼‰ã®ãƒãƒ¼ã‚«ãƒ«å±•é–‹ã‚’å®Ÿç¾ã™ã‚‹ãƒ•ãƒ¬ãƒ¼ãƒ ãƒ¯ãƒ¼ã‚¯ã‚’ç´¹ä»‹ã—ã¾ã™ã€‚ è¿‘å¹´ã€ã‚ªãƒ¼ãƒ—ãƒ³ã‚½ãƒ¼ã‚¹ã‚³ãƒŸãƒ¥ãƒ‹ãƒ†ã‚£ã‹ã‚‰ç™»å ´ã—ãŸå°åž‹é«˜æ€§èƒ½ãƒ¢ãƒ‡ãƒ«ã¨ã€ãƒ‡ãƒã‚¤ã‚¹æ€§èƒ½ã®å‘ä¸Šã«ã‚ˆã‚Šã€LLMã®åˆ©ç”¨å¯èƒ½æ€§ãŒå¤§ããæ‹¡å¤§ã—ã¾ã—ãŸã€‚ãã®ä¸ã§ã‚‚ãƒ–ãƒ©ã‚¦ã‚¶ã¯ã€è‡ªç„¶ãªã‚¨ãƒ¼ã‚¸ã‚§ãƒ³ãƒˆç’°å¢ƒã‚’æä¾›ã™ã‚‹ç†æƒ³çš„ãªãƒ—ãƒ©ãƒƒãƒˆãƒ•ã‚©ãƒ¼ãƒ ã¨ã—ã¦æ³¨ç›®ã•ã‚Œã¦ã„ã¾ã™ã€‚ èƒŒæ™¯ ã“ã‚Œã¾ã§ã®å¸¸è˜ã¨ã—ã¦ã€é«˜æ€§èƒ½ãªLLMã«ã¯é«˜ä¾¡ãªæ¥å‹™ç”¨ã‚°ãƒ©ãƒ•ã‚£ãƒƒã‚¯ã‚¹å‡¦ç†è£…ç½®ï¼ˆGPUï¼‰ãŒå¿…è¦ã§ã€ã‚¯ãƒ©ã‚¦ãƒ‰ã®ã‚µãƒ¼ãƒãƒ¼ã§ã—ã‹å‹•ä½œã•ã›ã‚‹ã“ã¨ãŒã§ãã¾ã›ã‚“ã§ã—ãŸã€‚ è¿‘å¹´ã€çŠ¶æ³ã¯å¤§ããå¤‰ã‚ã‚Šã¤ã¤ã‚ã‚Šã¾ã™ã€‚æ€§èƒ½ã‚’ç¶æŒã—ãŸã¾ã¾å°åž‹åŒ–ã•ã‚ŒãŸLLMãŒã€ã‚ªãƒ¼ãƒ—ãƒ³ã‚½ãƒ¼ã‚¹ã§å…¬é–‹ã•ã‚Œå§‹ã‚ãŸã®ã§ã™ã€‚ãƒ‘ãƒ©ãƒ¡ãƒ¼ã‚¿æ•°ï¼ˆãƒ¢ãƒ‡ãƒ«ã®è¤‡é›‘ã•ã‚’è¡¨ã™æŒ‡æ¨™ï¼‰ã¯10å„„ã‹ã‚‰30å„„ç¨‹åº¦ã«æŠ‘ãˆã‚‰ã‚Œã¦ã„ã¾ã™ã€‚ä¸€æ–¹ã§ã€ä¸€èˆ¬å‘ã‘ãƒ‘ã‚½ã‚³ãƒ³ã®å‡¦ç†èƒ½åŠ›ã‚‚è‘—ã—ãå‘ä¸Šã—ã¾ã—ãŸã€‚ä¾‹ãˆã°Appleã®æœ€æ–°ãƒ—ãƒã‚»ãƒƒã‚µM3ã‚’æè¼‰ã—ãŸ
deejayroka 2024/12/27
LLM

ãƒ¢ãƒ‡ãƒ«

é–‹ç™º

GPU
ãƒªãƒ³ã‚¯
æ–‡å—èµ·ã“ã—AIã®æ–°é¡”ã€ŒReazonSpeechã€ã€ãƒ†ãƒ¬ãƒ“ã§é›ãˆãŸéŸ³å£°èªè˜ã®ç²¾åº¦ã‚’è©¦ã—ãŸ
ç¶šã„ã¦ã€Colabã®ãƒ¡ãƒ‹ãƒ¥ãƒ¼ã®ã€Œãƒ©ãƒ³ã‚¿ã‚¤ãƒ ã€ã‹ã‚‰ã€Œãƒ©ãƒ³ã‚¿ã‚¤ãƒ ã®ã‚¿ã‚¤ãƒ—ã‚’å¤‰æ›´ã€ã‚’é¸æŠžã€‚åˆæœŸçŠ¶æ…‹ã§ã¯ã€Œãƒãƒ¼ãƒ‰ã‚¦ã‚§ã‚¢ã‚¢ã‚¯ã‚»ãƒ©ãƒ¬ãƒ¼ã‚¿ã€ã¯ã€ŒNoneã€ã¨ã„ã†è¨å®šã«ãªã£ã¦ã„ã‚‹ãŒã€ãƒ‡ãƒ¼ã‚¿å‡¦ç†ã‚’é«˜é€ŸåŒ–ã™ã‚‹ãŸã‚ã€ŒGPUã€ã«å¤‰æ›´ã—ã¦ä¿å˜ã—ãŸã€‚ã•ã‚‰ã«ã€æ¬¡ã®ã‚³ãƒ¼ãƒ‰ã§Google Driveã‚’ãƒžã‚¦ãƒ³ãƒˆã—ã€æ–‡å—èµ·ã“ã—ã‚’ã—ãŸã„éŸ³å£°ãƒ‡ãƒ¼ã‚¿ã€Œtest.m4aã€ã‚’ã‚¢ãƒƒãƒ—ãƒãƒ¼ãƒ‰ã—ã¦ãŠã„ãŸã€‚ ä½¿ç”¨ã—ãŸéŸ³å£°ãƒ‡ãƒ¼ã‚¿ã¯ã€æ—¥çµŒã‚¯ãƒã‚¹ãƒ†ãƒƒã‚¯ã«æŽ²è¼‰ã•ã‚Œã¦ã„ã‚‹ã‚¨ãƒ³ã‚¿ãƒ¼ãƒ—ãƒ©ã‚¤ã‚ºITåˆ†é‡Žã®ãƒ‹ãƒ¥ãƒ¼ã‚¹è¨˜äº‹ã‚’ã€ç†è€…ãŒèªã¿ä¸Šã’ã¦éŒ²éŸ³ã—ãŸã‚‚ã®ã ã€‚ãƒ¬ã‚¢ã‚¾ãƒ³ãƒ»ãƒ›ãƒ¼ãƒ«ãƒ‡ã‚£ãƒ³ã‚°ã‚¹ã«ã‚ˆã‚‹ã¨ç¾åœ¨å…¬é–‹ã—ã¦ã„ã‚‹å¦ç¿’æ¸ˆã¿ãƒ¢ãƒ‡ãƒ«ã¯ã€ReazonSpeechéŸ³å£°ã‚³ãƒ¼ãƒ‘ã‚¹ã®ä¸ã‹ã‚‰14ç§’ä»¥ä¸‹ã®ç™ºè©±ã«çµžã‚Šè¾¼ã‚“ã§å¦ç¿’ã•ã›ãŸã‚‚ã®ã ã¨ã„ã†ã€‚ãã“ã§ä»Šå›žã¯è¨˜äº‹ã®å†’é ã‚’ç´„14ç§’é–“ã ã‘èªã¿ä¸Šã’ã¦éŒ²éŸ³ã—ãŸã€‚ ã“ã“ã‹ã‚‰ã¯å®Ÿéš›ã®æ–‡å—èµ·ã“ã—ã®ãŸã‚ã®ä½œæ¥ã ã€‚ãƒ¬ã‚¢ã‚¾ãƒ³ãƒ»ãƒ›ãƒ¼ãƒ«ãƒ‡ã‚£ãƒ³ã‚°ã‚¹ãŒColabã§å…¬é–‹ã—ã¦ã„
deejayroka 2024/02/22
å¦ç¿’

GPU
ãƒªãƒ³ã‚¯
ãƒ¡ã‚¿ãŒæ±Žç”¨äººå·¥çŸ¥èƒ½ï¼ˆAGIï¼‰ã®ã‚ªãƒ¼ãƒ—ãƒ³ã‚½ãƒ¼ã‚¹åŒ–è¨ˆç”»ã€ã‚¶ãƒƒã‚«ãƒ¼ãƒãƒ¼ã‚°å®£è¨€ | Forbes JAPAN å…¬å¼ã‚µã‚¤ãƒˆï¼ˆãƒ•ã‚©ãƒ¼ãƒ–ã‚¹ ã‚¸ãƒ£ãƒ‘ãƒ³ï¼‰
ãƒ¡ã‚¿ã®ãƒžãƒ¼ã‚¯ãƒ»ã‚¶ãƒƒã‚«ãƒ¼ãƒãƒ¼ã‚°CEOã¯1æœˆ19æ—¥ã®ã‚¹ãƒ¬ãƒƒã‚ºã®æŠ•ç¨¿ã§ã€åŒç¤¾ãŒæ±Žç”¨äººå·¥çŸ¥èƒ½ï¼ˆAGIï¼‰ã®æ§‹ç¯‰ã«æ³¨åŠ›ã—ã€ã‚ªãƒ¼ãƒ—ãƒ³ã‚½ãƒ¼ã‚¹ã‚½ãƒ•ãƒˆã‚¦ã‚§ã‚¢ã¨ã—ã¦ä¸€èˆ¬å…¬é–‹ã™ã‚‹è¨ˆç”»ã‚’ã‚¢ãƒŠã‚¦ãƒ³ã‚¹ã—ãŸã€‚ ã€Œæ¬¡ä¸–ä»£ã®ã‚µãƒ¼ãƒ“ã‚¹ã®ãŸã‚ã«ã¯ã€AGIã®æ§‹ç¯‰ãŒå¿…é ˆã«ãªã‚‹ã“ã¨ãŒæ˜Žç¢ºã«ãªã‚Šã¤ã¤ã‚ã‚‹ã€ã¨ã€ã‚¶ãƒƒã‚«ãƒ¼ãƒãƒ¼ã‚°ã¯å€‹äººã§å…¬é–‹ã—ãŸå‹•ç”»ã®ä¸ã§èªžã‚Šã€æœ€é«˜ã®AIã‚¢ã‚·ã‚¹ã‚¿ãƒ³ãƒˆã‚„ã‚¯ãƒªã‚¨ã‚¤ã‚¿ãƒ¼ã¨ãƒ“ã‚¸ãƒã‚¹ã®ãŸã‚ã®AIã‚’æ§‹ç¯‰ã™ã‚‹ãŸã‚ã«ã¯ã€æŽ¨è«–ã‚„ã‚³ãƒ¼ãƒ‡ã‚£ãƒ³ã‚°ãªã©ã€ŒAIã®ã‚ã‚‰ã‚†ã‚‹å´é¢ã‚’é€²åŒ–ã•ã›ã‚‹ã“ã¨ãŒå¿…è¦ã ã€ã¨ä¸»å¼µã—ãŸã€‚ ã“ã®å–ã‚Šçµ„ã¿ã‚’ã‚µãƒãƒ¼ãƒˆã™ã‚‹ãŸã‚ã«ãƒ¡ã‚¿ã¯ã€2024å¹´æœ«ã¾ã§ã«ã‚¯ãƒ©ã‚¦ãƒ‰æ–½è¨ã«å·¨å¤§ãªã‚³ãƒ³ãƒ”ãƒ¥ãƒ¼ãƒˆãƒ‘ãƒ¯ãƒ¼ã‚’å°Žå…¥ã™ã‚‹è¨ˆç”»ã§ã€å…·ä½“çš„ã«ã¯ã€35ä¸‡åŸºã®ã‚¨ãƒŒãƒ“ãƒ‡ã‚£ã‚¢ã®H100ã€ã‚‚ã—ãã¯ç´„60ä¸‡åŸºã®ãã‚Œã¨åŒç‰ã®ãƒ‘ãƒ¯ãƒ¼ã‚’å®Ÿç¾ã™ã‚‹ä»–ç¤¾ã®GPUã‚’å°Žå…¥ã™ã‚‹äºˆå®šã¨ã„ã†ã€‚ ã“ã‚Œã»ã©ã®å°æ•°ã®H100ã‚’ç™ºæ³¨ã—ãŸä¼æ¥ã¯ãƒžã‚¤ã‚¯ãƒã‚½ãƒ•ãƒˆã®ã¿ã§ã‚ã‚Šã€ç´æœŸã¯1å¹´ã«ã‚‚ãŠã‚ˆã¶è¦‹
deejayroka 2024/01/22
â€œãƒ¡ã‚¿ã®ãƒžãƒ¼ã‚¯ãƒ»ã‚¶ãƒƒã‚«ãƒ¼ãƒãƒ¼ã‚°CEOã¯1æœˆ19æ—¥ã®ã‚¹ãƒ¬ãƒƒã‚ºã®æŠ•ç¨¿ã§ã€åŒç¤¾ãŒæ±Žç”¨äººå·¥çŸ¥èƒ½ï¼ˆAGIï¼‰ã®æ§‹ç¯‰ã«æ³¨åŠ›ã—ã€ã‚ªãƒ¼ãƒ—ãƒ³ã‚½ãƒ¼ã‚¹ã‚½ãƒ•ãƒˆã‚¦ã‚§ã‚¢ã¨ã—ã¦ä¸€èˆ¬å…¬é–‹ã™ã‚‹è¨ˆç”»ã‚’ã‚¢ãƒŠã‚¦ãƒ³ã‚¹ã—ãŸã€‚â€

ã‚¸ãƒ£ãƒ‘ãƒ³

ãƒ¢ãƒ‡ãƒ«

ãƒˆãƒ¬ãƒ¼ãƒ‹ãƒ³ã‚°

ã‚µã‚¤ãƒˆ

ai

GPU
ãƒªãƒ³ã‚¯
ç”ŸæˆAIå‘ã‘åŸºç›¤ã«135å„„å††ã‚’æŠ•ã˜ãŸç†ç”±ã¯ï¼Ÿã€€ã•ãã‚‰ã‚¤ãƒ³ã‚¿ãƒ¼ãƒãƒƒãƒˆã®ç”°ä¸ç¤¾é•·ã«èžã
å›½ç«‹èˆžé¶´å·¥æ¥é«˜ç‰å°‚é–€å¦æ ¡åœ¨å¦ä¸ã®1996å¹´ã«ã•ãã‚‰ã‚¤ãƒ³ã‚¿ãƒ¼ãƒãƒƒãƒˆã‚’å‰µæ¥ã—ã€ãƒ¬ãƒ³ã‚¿ãƒ«ã‚µãƒ¼ãƒãƒ¼äº‹æ¥ã‚’é–‹å§‹ã™ã‚‹ã€‚æœ€é«˜åŸ·è¡Œè²¬ä»»è€…ãªã©ã‚’æ´ä»»ã—ã€2007å¹´ã‚ˆã‚Šç¾è·ã€‚æ—¥æœ¬ãƒ‡ãƒ¼ã‚¿ã‚»ãƒ³ã‚¿ãƒ¼å”ä¼šã®ç†äº‹é•·ã‚„ã‚½ãƒ•ãƒˆã‚¦ã‚§ã‚¢å”ä¼šã®ä¼šé•·ãªã©ã‚’å‹™ã‚ã‚‹ã€‚1978å¹´ç”Ÿã¾ã‚Œã€‚å¤§é˜ªåºœå‡ºèº«ã€‚ï¼ˆå†™çœŸï¼šæ‘ç”° å’Œè¡ï¼‰ ç”ŸæˆAIï¼ˆäººå·¥çŸ¥èƒ½ï¼‰ã®é–‹ç™ºã«å‘ãé«˜æ€§èƒ½GPUåŸºç›¤ã«100å„„å††ä»¥ä¸Šã‚’æŠ•è³‡ã—ã¾ã™ã€‚ã©ã®ã‚ˆã†ãªçµŒå–¶åˆ¤æ–ãŒã‚ã£ãŸã®ã§ã—ã‚‡ã†ã€‚ ç´„135å„„å††ã‚’æŠ•è³‡ã—ã¦ã€ç±³ã‚¨ãƒŒãƒ“ãƒ‡ã‚£ã‚¢ã®GPUã€ŒNVIDIA H100 Tensor ã‚³ã‚¢ GPUã€ã‚’2000åŸºä»¥ä¸Šå°Žå…¥ã—ã¦ã‚¤ãƒ³ãƒ•ãƒ©ã‚’æ•´å‚™ã—ã¾ã™ï¼ˆæœ¬èªŒæ³¨ï¼šã†ã¡ç´„68å„„å††ã¯çµŒæ¸ˆç”£æ¥çœã‹ã‚‰è£œåŠ©ã‚’å—ã‘ã‚‹ï¼‰ã€‚2024å¹´1æœˆä»¥é™ã‚¯ãƒ©ã‚¦ãƒ‰ã‚µãƒ¼ãƒ“ã‚¹ã¨ã—ã¦æä¾›ã§ãã‚‹äºˆå®šã§ã™ã€‚ å®Ÿã¯2020å¹´ã”ã‚ã«AIå‘ã‘é«˜æ€§èƒ½GPUã¸ã®æŠ•è³‡ã‚’ã„ã£ãŸã‚“è¦‹åˆã‚ã›ã‚‹åˆ¤æ–ã‚’ã—ã¦ã„ã¾ã—ãŸã€‚GPUã®ä¾¡æ ¼ãŒé«˜é¨°ã—ã‚°ãƒãƒ¼ãƒãƒ«ãªèª¿é”ç«¶äº‰ãŒåŽ³ã—ããªã£ãŸã“ã¨
deejayroka 2023/11/09
â€œç”ŸæˆAIï¼ˆäººå·¥çŸ¥èƒ½ï¼‰ã®é–‹ç™ºã«å‘ãæœ€å…ˆç«¯ã‚¯ãƒ©ã‚¦ãƒ‰ã¯ç¾æ™‚ç‚¹ã§å›½å†…å”¯ä¸€ã â€

ai

é–‹ç™º

æŠ•è³‡

GPU

ã‚¤ãƒ³ã‚¿ãƒ¼ãƒãƒƒãƒˆ
ãƒªãƒ³ã‚¯
NVIDIAã€Stable Diffusionã‚’å€é€Ÿã«ã™ã‚‹TensorRTã€‚æ–°ãƒ‰ãƒ©ã‚¤ãƒã§å¯¾å¿œ
deejayroka 2023/10/19
â€œæœ¬ãƒ‰ãƒ©ã‚¤ãƒã§ã¯ã€WebUIã€ŒAutomatic1111ã€ç‰ˆã®Stable Diffusionã®å‡¦ç†é€Ÿåº¦ã‚’å€é€Ÿã«ã™ã‚‹TensorRTã‚¢ã‚¯ã‚»ãƒ©ãƒ¬ãƒ¼ã‚·ãƒ§ãƒ³ã‚’è¿½åŠ â€

ai

GPU

Stable-Diffusion

nvidia

videocard

StableDiffusion

GeForce

SD

stable_diffusion
ãƒªãƒ³ã‚¯
Meteor Lakeã®å†…è”µNPUã€AIå‡¦ç†ã®é›»åŠ›åŠ¹çŽ‡ã¯CPUã®7.8å€
deejayroka 2023/09/20
â€œNPUã¯ã€åŒç¤¾ãŒè²·åŽã—ãŸMovidiusãŒé–‹ç™ºã—ã¦ããŸVPUãŒãƒ™ãƒ¼ã‚¹ã«ãªã£ã¦ãŠã‚Šã€ä½Žæ¶ˆè²»é›»åŠ›ã§AIæŽ¨è«–ãŒå¯èƒ½ã«ãªã‚‹ãªã©ã®ç‰¹å¾´ã‚’å‚™ãˆã¦ã„ã‚‹â€

CPU

GPU

ai
ãƒªãƒ³ã‚¯
NVIDIAã®æ¬¡ä¸–ä»£GPUã€ŒBlackwellã€ã¯ã€NVIDIAå²ä¸Šæœ€å¤§ã®æ€§èƒ½å‘ä¸Šã«ãªã‚‹å¯èƒ½æ€§ | TEXAL
NVIDIAã®GPUã¯ã€ã¾ã RTX 4000ã‚·ãƒªãƒ¼ã‚ºãŒç™ºå£²ã•ã‚ŒãŸã°ã‹ã‚Šã ãŒã€æ—¢ã«æ¬¡ä¸–ä»£GPUã®æ°—ã«ãªã‚‹å™‚ãŒå‡ºå§‹ã‚ã¦ã„ã‚‹ã€‚ãã‚Œã«ã‚ˆã‚‹ã¨ã€åŒç¤¾ã®æ¬¡æœŸã€ŒBlackwellã€ã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã§ã¯ã€åŒç¤¾åˆã®Mult-Chip Moduleï¼ˆMCMï¼‰ãŒæŽ¡ç”¨ã•ã‚Œã€NVIDIAã®GPUã®æ´å²ã®ä¸ã§ã€æœ€å¤§ã®æ€§èƒ½å‘ä¸Šã‚’å®Ÿç¾ã™ã‚‹å¯èƒ½æ€§ãŒéžå¸¸ã«é«˜ã„ã¨ã®ã“ã¨ã ã€‚ Blackwellã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã¯ã€ç¾è¡Œã®Hopperã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã®å¾Œç¶™ã¨ã—ã¦é–‹ç™ºãŒé€²ã‚ã‚‰ã‚Œã¦ã„ã‚‹ã‚‚ã®ã¨è¨€ã‚ã‚Œã¦ã„ã‚‹ã€‚Hopperã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ã®ã‚³ãƒ³ã‚·ãƒ¥ãƒ¼ãƒžå‘ã‘ã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£ãŒã€ŒAda Lovelaceã€ã§ã‚ã£ãŸäº‹ã‹ã‚‰ã€Blackwellã«ã‚‚ã‚³ãƒ³ã‚·ãƒ¥ãƒ¼ãƒžã‚»ã‚°ãƒ¡ãƒ³ãƒˆå‘ã‘ã®å¯¾å¿œè£½å“ãŒåˆ¥ã®åã§ç”¨æ„ã•ã‚Œã¦ã„ã‚‹å¯èƒ½æ€§ãŒã‚ã‚‹ã€‚ã‚ã‚‹ã„ã¯ã€BlackwellãŒã‚³ãƒ³ã‚·ãƒ¥ãƒ¼ãƒžå‘ã‘ã®åå‰ã§ã€ãƒ‡ãƒ¼ã‚¿ã‚»ãƒ³ã‚¿ãƒ¼å‘ã‘ãŒåˆ¥ã®ã‚³ãƒ¼ãƒ‰ãƒãƒ¼ãƒ ã§ã‚ã‚‹å¯èƒ½æ€§ã‚‚ã‚ã‚‹ã€‚ã„ãšã‚Œã«ã›ã‚ˆã€Blackwel
deejayroka 2023/09/20
GPU

ã‚¢ãƒ¼ã‚ãƒ†ã‚¯ãƒãƒ£
ãƒªãƒ³ã‚¯
ChatGPTã®ã‚ˆã†ãªãƒ¢ãƒ‡ãƒ«ã‚’é«˜é€Ÿç”Ÿæˆã™ã‚‹ãƒ„ãƒ¼ãƒ«ã€MicrosoftãŒé–‹ç™ºã€€ã€Œæ—¢å˜æŠ€è¡“ã®15å€ä»¥ä¸Šã€ã†ãŸã†
ç±³Microsoftã¯4æœˆ12æ—¥ã€ChatGPTã®ã‚ˆã†ãªãƒ¢ãƒ‡ãƒ«ã‚’ç”Ÿæˆã§ãã‚‹ãƒ„ãƒ¼ãƒ«ã€ŒDeepSpeed-Chatã€ã‚’å…¬é–‹ã—ãŸã€‚GPU1å°ã§100å„„ä»¥ä¸Šã®ãƒ‘ãƒ©ãƒ¡ãƒ¼ã‚¿ã‚’æŒã¤ãƒ¢ãƒ‡ãƒ«ã®å¦ç¿’ãŒã§ãã‚‹ä¸Šã€å¦ç¿’é€Ÿåº¦ã‚‚æ—¢å˜æŠ€è¡“ã®15å€ä»¥ä¸Šã¨ã—ã¦ã„ã‚‹ã€‚ DeepSpeed-Chatã¯ChatGPTã§ã‚‚æŽ¡ç”¨ã•ã‚Œã¦ã„ã‚‹å¦ç¿’æ‰‹æ³•ã€ŒRLHFã€ã‚’åŠ¹çŽ‡çš„ã«å®Ÿè¡Œã™ã‚‹ãŸã‚ã®ãƒ„ãƒ¼ãƒ«ã€‚ã“ã‚Œã¾ã§ã¯é«˜ä¾¡ãªGPUã‚’å¤§é‡ã«ä½¿ã†å¿…è¦ãŒã‚ã£ãŸãŒã€DeepSpeed-Chatã¯å°‘ãªã„GPUã§åŠ¹çŽ‡çš„ã«å¦ç¿’ãŒã§ãã‚‹ã€‚ ä¾‹ãˆã°ã€ã‚¯ãƒ©ã‚¦ãƒ‰ä¸Šã§GPUã€ŒA100-80GB x8ã€ã‚’ä½¿ã£ã¦å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒOPT-B13ã€ã‚’è¨“ç·´ã™ã‚‹å ´åˆã€9æ™‚é–“ç¨‹åº¦ã§å®Œäº†ã—ã€ã‚³ã‚¹ãƒˆã‚‚290ãƒ‰ãƒ«ç¨‹åº¦ã¨ã„ã†ã€‚è¤‡æ•°å°ã®GPUã‚’ä½¿ãˆã°1000å„„ãƒ‘ãƒ©ãƒ¡ãƒ¼ã‚¿è¦æ¨¡ã®ãƒ¢ãƒ‡ãƒ«ã®å¦ç¿’ã«ã‚‚å¯¾å¿œã™ã‚‹ã€‚ é–¢é€£è¨˜äº‹ ChatGPTãŒè·å‹™çµŒæ´æ›¸ã‚’ä½œæˆã™ã‚‹ã‚µãƒ¼ãƒ“ã‚¹ç¶šã€…ã€€ã‚¨ãƒ³ã‚¸ãƒ‹ã‚¢å‘ã‘ã‹ã‚‰çœ‹è·å¸«ã¾ã§ ãƒ•ã‚¡ã‚¤
deejayroka 2023/04/19
â€œ1ï½ž2æ™‚é–“ã®ã‚³ãƒ¼ãƒ’ãƒ¼ã‚¿ã‚¤ãƒ ã‚„æ˜¼ä¼‘ã¿ã«ã€DeepSpeed-Chatã§å°è¦æ¨¡ãªãƒˆã‚¤ãƒ¢ãƒ‡ãƒ«ã‚’ãƒˆãƒ¬ãƒ¼ãƒ‹ãƒ³ã‚°ã—ã¦ã¿ã‚‹ã®ã‚‚è‰¯ã„ã§ã—ã‚‡ã†â€

ChatGPT

ãƒ¢ãƒ‡ãƒ«

ãƒ„ãƒ¼ãƒ«

å¦ç¿’

æŠ€è¡“

é–‹ç™º

Microsoft

GPU

technology

tool
ãƒªãƒ³ã‚¯
Metaã®å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒLLaMAã€ã‚’M1æè¼‰Macã§å®Ÿè¡Œå¯èƒ½ã«ã—ãŸã€Œllama.cppã€ãŒã‚¢ãƒƒãƒ—ãƒ‡ãƒ¼ãƒˆã«ã‚ˆã‚Šã‚ãšã‹6GBæœªæº€ã®ãƒ¡ãƒ¢ãƒªä½¿ç”¨é‡ã§å®Ÿè¡Œå¯èƒ½ã« - GIGAZINE
2023å¹´2æœˆã«MetaãŒç™ºè¡¨ã—ãŸå¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒLLaMAã€ã¯ã€å¾“æ¥ã®GPT-3ã‚ˆã‚Šã‚‚å°è¦æ¨¡ã§ã‚ã‚ŠãªãŒã‚‰GPT-3ã«åŒ¹æ•µã™ã‚‹æ€§èƒ½ã‚’å˜ä½“GPUã®ç’°å¢ƒã§ã‚‚ç¤ºã™ã“ã¨ãŒå¯èƒ½ã¨ã•ã‚Œã¦ãŠã‚Šã€2023å¹´3æœˆã«ã¯ã‚¨ãƒ³ã‚¸ãƒ‹ã‚¢ã®ã‚¸ãƒ§ãƒ¼ã‚¸ãƒ»ã‚²ãƒ«ã‚¬ãƒŽãƒ•æ°ãŒM1ãªã©ã®Apple ã‚·ãƒªã‚³ãƒ³æè¼‰Macã§LLaMAã‚’å‹•ä½œã•ã›ã‚‹ã€Œllama.cppã€ã‚’å…¬é–‹ã—ã¾ã—ãŸã€‚ãã‚“ãªä¸ã€ãƒ—ãƒã‚°ãƒ©ãƒžãƒ¼ã®ã‚¸ãƒ£ã‚¹ãƒ†ã‚£ãƒ³ãƒ»ã‚¿ãƒ‹ãƒ¼æ°ãŒllama.cppãŒå‹•ä½œã™ã‚‹éš›ã®ãƒ¡ãƒ¢ãƒªä½¿ç”¨é‡ã‚’æ¸›ã‚‰ã™ã‚¢ãƒƒãƒ—ãƒ‡ãƒ¼ãƒˆã‚’è¡Œã„ã€LLaMAã®ä¸€éƒ¨ãƒ¢ãƒ‡ãƒ«ã«è‡³ã£ã¦ã¯6GBæœªæº€ã®RAMã§å‹•ä½œã™ã‚‹ã“ã¨ãŒå ±å‘Šã•ã‚Œã¦ã„ã¾ã™ã€‚ Make loading weights 10-100x faster by jart Â· Pull Request #613 Â· ggerganov/llama.cpp Â· GitHub https://github.com/ggerganov/llam
deejayroka 2023/04/05
â€œ2023å¹´3æœˆ31æ—¥ã«ã‚¿ãƒ‹ãƒ¼æ°ã¯llama.cppã®C++ã‚½ãƒ¼ã‚¹ã‚³ãƒ¼ãƒ‰ã«ã‚¢ãƒƒãƒ—ãƒ‡ãƒ¼ãƒˆã‚’åŠ ãˆãŸã“ã¨ã‚’å ±å‘Šã—ã¾ã—ãŸã€‚ã‚¿ãƒ‹ãƒ¼æ°ã«ã‚ˆã‚‹ã‚¢ãƒƒãƒ—ãƒ‡ãƒ¼ãƒˆã®çµæžœã€LLaMAã®å®Ÿè¡Œã®éš›ã®ãƒ¡ãƒ¢ãƒªä½¿ç”¨é‡ãŒå¤§å¹…ã«æ¸›å°‘ã—ã€å¾“æ¥ã¯30GBå¿…è¦ã ã£ãŸLLaMAã®13Bãƒ¢ãƒ‡ãƒ«ã®

mac

ã‚¢ãƒƒãƒ—ãƒ‡ãƒ¼ãƒˆ

GPU

github

language
ãƒªãƒ³ã‚¯
DeepSpeedã‚’ç”¨ã„ãŸHuggingface Transformersã®è¤‡æ•°ãƒŽãƒ¼ãƒ‰ã§ã®å¦ç¿’ - Retrieva TECH BLOG
deejayroka 2023/02/07
"ZeROãªã©ã®çœãƒ¡ãƒ¢ãƒªåŒ–ã‚„16bitæµ®å‹•å°æ•°ç‚¹ã‚’ç”¨ã„ãŸé«˜é€ŸåŒ–ãªã©ã®æ©Ÿèƒ½ã¨çµ„ã¿åˆã‚ã›ã¦ç”¨ã„ã‚‹ã“ã¨ã‚‚å¯èƒ½ãªã®ã§ã€Huggingface Transformersã®ã¿ã®æ™‚ã‚ˆã‚Šã‚‚å¤§ããªãƒ‘ãƒ©ãƒ¡ãƒ¼ã‚¿ã‚µã‚¤ã‚ºã®ãƒ¢ãƒ‡ãƒ«ã®å¦ç¿’ã‚’è¡Œã†ã“ã¨ãŒå¯èƒ½ã«ãªã‚Šã¾ã™"

å¦ç¿’

ãƒ¢ãƒ‡ãƒ«

GPU
ãƒªãƒ³ã‚¯
Vertexã§3ãƒ¶æœˆã§ä½œã‚‹é‹ç”¨å¯èƒ½ãªML APIåŸºç›¤ - CADDi Tech Blog
ã“ã‚“ã«ã¡ã¯ã€‚CADDiã®AI Labã§MLOpsã‚¨ãƒ³ã‚¸ãƒ‹ã‚¢ã‚’ã‚„ã£ã¦ã„ã‚‹ä¸æ‘éµä»‹ã§ã™ã€‚ MLOpsãƒãƒ¼ãƒ ã¯ä»Šã‹ã‚‰3ãƒ¶æœˆå‰ã«ç«‹ã¡ä¸ŠãŒã£ãŸã°ã‹ã‚Šã®æ–°ã—ã„ãƒãƒ¼ãƒ ãªã®ã§ã™ãŒã€ãã®å‰èº«ã¨ã—ã¦APIåŸºç›¤ã‚’ä½œã£ã¦ã„ãŸæ™‚æœŸãŒã‚ã£ãŸã®ã§ã€ãã“ã§å¾—ã‚‰ã‚ŒãŸçŸ¥è¦‹ã‚’æ›¸ã„ã¦ã„ã“ã†ã¨æ€ã„ã¾ã™ã€‚ èƒŒæ™¯ CADDiã®AI Labã¯2021å¹´ã®12æœˆã«ç«‹ã¡ä¸ŠãŒã£ãŸä»Šæœˆ1æ‰ã«ãªã£ãŸã°ã‹ã‚Šã®çµ„ç¹”ã§ã™ã€‚ãã®è‹¥ã•ã«ã‚‚é–¢ã‚ã‚‰ãšã€æ—¥ã€…æœ‰ç”¨ãªMLãƒ¢ãƒ‡ãƒ«ãŒä½œæˆã•ã‚Œã¦ã„ã£ã¦ã„ã¾ã™ã€‚ ãã®ã‚ˆã†ãªä¸ã§ã€ã€Œæ–°ã—ãä½œã£ãŸMLãƒ¢ãƒ‡ãƒ«ã‚’ç´ æ—©ããƒ¦ãƒ¼ã‚¶ã«ãƒ‡ãƒªãƒãƒªãƒ¼ã—ãŸã„ã€ã¨ã„ã†è©±ãŒä¸ŠãŒã‚‹ã‚ˆã†ã«ãªã‚Šã¾ã—ãŸã€‚ã“ã“ã§ã„ã†ãƒ¦ãƒ¼ã‚¶ã¨ã¯CADDiç¤¾å“¡ã‚„ç¤¾å†…ã‚·ã‚¹ãƒ†ãƒ ã€å…¬é–‹ã‚¢ãƒ—ãƒªã‚±ãƒ¼ã‚·ãƒ§ãƒ³ãªã©ã‚’æŒ‡ã—ã¾ã™ã€‚ ãã®ãŸã‚ã€AI Labå†…ã§ç°¡å˜ã«ä½¿ç”¨ã§ãã‚‹APIåŸºç›¤ã‚’ä½œæˆã™ã‚‹ã“ã¨ã«ã—ã¾ã—ãŸã€‚å…·ä½“çš„ã«ã¯ä»¥ä¸‹ã®ä½“é¨“ã‚’ä½œã‚‹ã“ã¨ã‚’ç›®æŒ‡ã—ã¾ã—ãŸã€‚ é–‹ç™ºè€…ã«æä¾›ã™ã‚‹APIãƒ‡ãƒ—ãƒã‚¤ä½“é¨“ æŽ¨è«–ã‚³ãƒ¼ãƒ‰
deejayroka 2023/01/22
ã‚ã¨ã§èªã‚€

AI

API

ML

ãƒ¢ãƒ‡ãƒ«

ã‚µãƒ¼ãƒ

ãƒ†ã‚¹ãƒˆ

ai

tech

GPU
ãƒªãƒ³ã‚¯
ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã®æŽ¨è«–æ™‚é–“ã‚’å¤§å¹…ã«çŸç¸®ã§ãã‚‹ã€ŒNVIDIA TensorRTã€åˆ©ç”¨æ–¹æ³•ã¯ï¼Ÿ | Ledge.ai
ã‚µã‚¤ãƒ³ã‚¤ãƒ³ã—ãŸçŠ¶æ…‹ã§ã€Œã„ã„ãã€ã‚’æŠ¼ã™ã¨ã€ãƒžã‚¤ãƒšãƒ¼ã‚¸ã® ã€Œã„ã„ãå±¥æ´ã€ã«ä¸€è¦§ã¨ã—ã¦ä¿å˜ã•ã‚Œã¦ã„ãã®ã§ã€ å†åº¦èªã¿ãŸããªã£ãŸæ™‚ã‚„ã€ã‚ã¨ã§ã˜ã£ãã‚Šèªã¿ãŸã„ã¨ãã«ä¾¿åˆ©ã§ã™ã€‚
deejayroka 2023/01/16
ãƒ¢ãƒ‡ãƒ«

ai

ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°

TensorFlow

GPU

deep learning

ML

æ™‚é–“
ãƒªãƒ³ã‚¯
The Technology Behind BLOOM Training
Please note that both Megatron-LM and DeepSpeed have Pipeline Parallelism and BF16 Optimizer implementations, but we used the ones from DeepSpeed as they are integrated with ZeRO. Megatron-DeepSpeed implements 3D Parallelism to allow huge models to train in a very efficient way. Letâ€™s briefly discuss the 3D components. DataParallel (DP) - the same setup is replicated multiple times, and each being
deejayroka 2022/08/16
GPU

technology

deeplearning
ãƒªãƒ³ã‚¯
AMD Ryzenè¨è¨ˆè€…ã®å¤©æ‰ã‚¸ãƒ ãƒ»ã‚±ãƒ©ãƒ¼ãŒãƒ†ã‚¹ãƒ©ã‚’è¾žã‚ã¦Intelã¸é›»æ’ƒç§»ç±
AMDã§Athlon(K7)ã‚„Ryzenã®é–‹ç™ºã‚’ä¸»å°Žã—ãŸå¤©æ‰ãƒ—ãƒã‚»ãƒƒã‚µãƒ‡ã‚¶ã‚¤ãƒŠãƒ¼ã®ã‚¸ãƒ ãƒ»ã‚±ãƒ©ãƒ¼æ°ãŒã€ãƒ†ã‚¹ãƒ©ã‚’è¾žã‚ã¦Intelã«é›»æ’ƒç§»ç±ã—ãŸã“ã¨ãŒæ˜Žã‚‰ã‹ã«ãªã‚Šã¾ã—ãŸã€‚ Jim Keller Joins Intel to Lead Silicon Engineering | Intel Newsroom https://newsroom.intel.com/news-releases/jim-keller-joins-intel-lead-silicon-engineering/ DECã€AMDã€Appleãªã©ã‚’çµŒã¦ããŸã‚±ãƒ©ãƒ¼æ°ã¯ã€AMDã§Athlon(K7)ã‚„Athlon 64ã‚’é–‹ç™ºã—ã¦Intelã‚’å€’ã—ãŸã“ã¨ã§çŸ¥ã‚‰ã‚Œã‚‹ã€ãƒ—ãƒã‚»ãƒƒã‚µè¨è¨ˆã§ã¯å¤©æ‰ã¨ã®å‘¼ã³å£°ãŒé«˜ã„äººç‰©ã§ã™ã€‚ã‚±ãƒ©ãƒ¼æ°ã¯Appleã§ã¯A4ã‚„A5ã®è¨è¨ˆã‚„MacBook Airã®é–‹ç™ºã«é–¢ã‚ã£ãŸã“ã¨ã§ã‚‚çŸ¥ã‚‰ã‚Œã¦ã„ã¾ã™ã€‚ 2012å¹´ã«å¤å·£AM
deejayroka 2018/04/27
é–‹ç™º

GPU

CPU

Intel

AMD
ãƒªãƒ³ã‚¯
ä¸–ç•Œã‚’å¤‰ãˆã‚‹AIæŠ€è¡“ã€Œãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã€ãŒè£½é€ æ¥ã«ã‚‚ãŸã‚‰ã™ã‚¤ãƒ³ãƒ‘ã‚¯ãƒˆ
å¦ç¿’ã—ç¶šã‘ã‚‹äººå·¥çŸ¥èƒ½ã®ã‚µã‚¤ã‚¯ãƒ« å¦ç¿’å‡¦ç†ã¨æŽ¨è«–å‡¦ç†ã‚’åŒä¸€ã®æ©Ÿå™¨ã§è¡Œã†å¿…è¦ã¯ã‚ã‚Šã¾ã›ã‚“ã€‚å¤§é‡ã®ãƒ‡ãƒ¼ã‚¿èªã¿è¾¼ã¾ã›ã‚‹å¿…è¦ã®ã‚ã‚‹å¦ç¿’å‡¦ç†ã¯é«˜æ€§èƒ½ãªGPUã‚µãƒ¼ãƒä¸Šã§è¡Œã„ã€ãã®çµæžœã€ä½œè£½ã•ã‚ŒãŸå¦ç¿’æ¸ˆã¿ãƒ¢ãƒ‡ãƒ«ã‚’åˆ©ç”¨ã—ã¦çµ„ã¿è¾¼ã¿æ©Ÿå™¨ã«æè¼‰ã•ã‚ŒãŸJetson TX1ä¸Šã§æŽ¨è«–å‡¦ç†ã‚’å®Ÿè¡Œã™ã‚‹ã¨ã„ã†å®Ÿè£…ã‚‚å¯èƒ½ã¨ãªã‚Šã¾ã™ã€‚ã“ã“ã§ã¯ã€ã“ã®ãƒ•ãƒãƒ¼ã‹ã‚‰å°Žã‘ã‚‹æ´»ç”¨æ³•ã«ã¤ã„ã¦èª¬æ˜Žã—ã¾ã™ã€‚ ä¾‹ãˆã°ã€ã•ã¾ã–ã¾ãªè£½å“ç”»åƒã‚’èªè˜ã§ãã‚‹ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚’å¦ç¿’ã•ã›ã¦ã€å€‰åº«ãƒãƒœãƒƒãƒˆã‚’ä½œã‚‹ã“ã¨ã‚’è€ƒãˆã¦ã¿ã¾ã—ã‚‡ã†ã€‚ å®Œæˆã—ãŸã“ã®ãƒãƒœãƒƒãƒˆã¯ã€è‡ªå¾‹çš„ã«å€‰åº«ã‚’ç§»å‹•ã—ã€æ£šã‚„ç®±ã®ä¸ã®å„è£½å“ã‚’è˜åˆ¥ã™ã‚‹ã“ã¨ãŒã§ãã¾ã™ã€‚ã—ã‹ã—ã‚ã‚‹æ—¥ã€ä»Šã¾ã§æ‰±ã£ã¦ã„ãŸè£½å“ã¨ã¯å…¨ãè¦‹ãŸç›®ã®ç•°ãªã‚‹è£½å“ãŒç™»éŒ²ã•ã‚Œã¦å€‰åº«å†…ã§æ‰±ã‚ã‚Œã‚‹ã‚ˆã†ã«ãªã£ãŸã¨ã—ã¾ã—ã‚‡ã†ã€‚ ãƒãƒœãƒƒãƒˆã¯ã“ã®è£½å“ã‚’æ‰±ã†ã“ã¨ãŒå‡ºæ¥ãªããªã£ã¦ã—ã¾ã„ã¾ã™ãŒã€ã“ã“ã§è§£æ±ºæ–¹æ³•ãŒã‚ã‚Šã¾ã™ã€‚èªè˜ã§ããªã‹ã£ãŸè£½å“ã®ã‚«ãƒ¡ãƒ©ç”»åƒã‚’ã‚µ
deejayroka 2018/04/02
ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯

ãƒãƒœãƒƒãƒˆ

GPU

ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°

ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼

ãƒ•ã‚¡ãƒŠãƒƒã‚¯
ãƒªãƒ³ã‚¯
ãã‚‚ãã‚‚AIã§ä½•ãŒã§ãã‚‹ã®ã‹ï¼Ÿå®Ÿç¾å¯èƒ½ãª7ã¤ã®æˆæžœ | æ ªå¼ä¼šç¤¾ã‚¤ãƒ³ã‚ãƒ¥ãƒ“ãƒƒãƒˆ
ã€ŒAIï¼ˆäººå·¥çŸ¥èƒ½ï¼‰ã‚’æ´»ç”¨ã—ã¦ãƒ“ã‚¸ãƒã‚¹ã§æˆæžœã‚’ã‚ã’ã‚ˆã†ã€ã¨ã„ã†å‹•ããŒã¾ã™ã¾ã™é«˜ã¾ã£ã¦ãã¾ã—ãŸã€‚ ã—ã‹ã—ä¸€æ–¹ã§ã€ŒAIã‚’é”æ³•ã®æ–ã ã¨èª¤è§£ã—ãŸäººãŸã¡ãŒã€ãƒ ãƒãƒ£ãªè¦æœ›ã‚’å‡ºã—ã¦ãã‚‹ã€ã¨ã„ã†ã‚ˆã†ãªãƒœãƒ¤ã‚ã‚‚ã€ã¾ãŸã‚ˆãè€³ã«ã™ã‚‹è©±ã§ã™ã€‚ ã¤ã¾ã‚ŠAIé–¢é€£ã®æŠ€è¡“ã«ã‚ˆã£ã¦ã€ä½•ãŒã§ãã¦ä½•ãŒã§ããªã„ã®ã‹ï¼Ÿã¨ã„ã†ç‚¹ãŒã‚ã„ã¾ã„ãªã¾ã¾ã«ã€æœŸå¾…ã ã‘ãŒå…ˆè¡Œã—ãŒã¡ã¨ã„ã†ã®ãŒå¤§æ–¹ã®ç¾çŠ¶ã¨ã„ãˆãã†ã§ã™ã€‚ ãã‚“ãªä¸ã§ã¡ã‚‡ã£ã¨ä¾¿åˆ©ãªå›³ã‚’ã¿ã¤ã‘ã¾ã—ãŸï¼ˆè¨˜äº‹æœ€ä¸Šéƒ¨ã€‚ã‚ªãƒªã‚¸ãƒŠãƒ«ã‚’ã‚‚ã¨ã«AI4Uç·¨é›†éƒ¨ã§ä½œæˆï¼‰ã€‚ ã€ŒAIã«ã‚ˆã‚‹ï¼—ã¤ã®æˆæžœã€ï¼ˆSeven spectrum of outcomes for AIï¼‰ã¨é¡Œã•ã‚ŒãŸå›³ã€‚ãã®åã®é€šã‚Šã€AIã«ã‚ˆã£ã¦è§£æ±ºã§ãã‚‹æˆæžœã€ã¤ã¾ã‚Šãƒ¦ãƒ¼ã‚¶ãƒ¼ãƒ‹ãƒ¼ã‚ºã‚’ï¼—æ®µéšŽã§æ•´ç†ã—ã¦ã„ã¾ã™ã€‚ ã€ŒèªçŸ¥ã€ã‚„ã€Œé€šçŸ¥ã€ã®ã‚ˆã†ã«ç¾æ™‚ç‚¹ã®æŠ€è¡“ãƒ¬ãƒ™ãƒ«ã§å¯èƒ½ãªæ®µéšŽã‚‚ã‚ã‚Œã°ã€äººã®åˆ¤æ–ã‚’æ‰‹åŠ©ã‘ã™ã‚‹ã€Œç’°å¢ƒèªçŸ¥ã€ã¨ã„ã£ãŸã¾ã é›£ã—ã„ãƒ¬ãƒ™ãƒ«ã‚‚ã‚ã‚Šã¾ã™ã€‚ AIäº‹
deejayroka 2017/08/19
AI

GPU

ã‚µã‚¤ãƒãƒ¼ã‚¨ãƒ¼ã‚¸ã‚§ãƒ³ãƒˆ

ãƒãƒ¼ãƒ‰ã‚¦ã‚§ã‚¢

Deep learning

Microsoft

ãƒªã‚¯ãƒ«ãƒ¼ãƒˆ

ãƒ”ã‚¯ã‚»ãƒ«

ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°

ãƒˆãƒ©ãƒ³ã‚¹ã‚³ã‚¹ãƒ¢ã‚¹
ãƒªãƒ³ã‚¯
https://jp.techcrunch.com/2017/06/09/20170608facebook-is-speeding-up-training-for-visual-recognition-models/
deejayroka 2017/06/09
GPU

æ™‚é–“

Facebook

ãƒ¢ãƒ‡ãƒ«
ãƒªãƒ³ã‚¯
1