Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Samba Cloudの高速推論を活用した模範解答分析と開発知見
Search
いおりん
October 08, 2025
0
160
Samba Cloudの高速推論を活用した模範解答分析と開発知見
いおりん
October 08, 2025
Tweet
Share
More Decks by いおりん
See All by いおりん
claude codeでPrompt Engineering
iori0311
0
790
Featured
See All Featured
The Pragmatic Product Professional
lauravandoore
37
7.1k
Code Review Best Practice
trishagee
74
19k
Collaborative Software Design: How to facilitate domain modelling decisions
baasie
0
110
How Fast Is Fast Enough? [PerfNow 2025]
tammyeverts
3
420
Practical Orchestrator
shlominoach
190
11k
Building the Perfect Custom Keyboard
takai
1
670
Designing for Performance
lara
610
70k
It's Worth the Effort
3n
187
29k
Game over? The fight for quality and originality in the time of robots
wayneb77
1
72
SEO in 2025: How to Prepare for the Future of Search
ipullrank
3
3.3k
AI: The stuff that nobody shows you
jnunemaker
PRO
1
37
Balancing Empowerment & Direction
lara
5
830
Transcript
4BNCB$MPVEͷߴਪΛ ׆༻ͨ͠ൣղੳͱ։ൃݟ Samba Cloud Tokyo Mingle 4 202510݄7 גࣜձࣾ ϚΠΫϩγϛϡϨʔγϣϯ
ࣗݾհ גࣜձࣾϚΠΫϩγϛϡϨʔγϣϯ ٱอ ҏ৫ σδλϧ࠾αʔϏεtestusͷ։ൃ
ൣճ ࠾ج४ LLMʹΑΔ࠾ LLMΛ׆༻ͨ͠࠾ϑϩʔ
࠾ج४ੜͷͨΊͷൣղੳͱϞσϧͷੑೳ ൣղ: εϙʔπΛ͢Δ͜ͱɺָ͍͚ͩ͠Ͱͳ͘ɺ݈߁ʹྑ͍ɻ ## ൣճͷྨ ɾӳจ༁ ɾจӳ༁ ɾࣗ༝ӳ࡞ ## ൣղͷੳ
ΠσΟΦϜߏจ จ๏ ## ࠾ج४ͷੜ Prompt gemini-2.5-flash( ⚪︎ ) gemini-2.5-flash-lite(×) ӳ༁ͱͯ͠ͷ ج४ʹͳ͍ͬͯΔ
ίετͱੜ࣌ؒ ※1 τʔΫϯྔͱੜ࣌ؒGoogle AI studioΛ༻ ※2 ίετܭࢉVertex AIͷྉۚදΑΓࢉग़ ※3 ίϯςΩετΩϟογϡͷ༻ͳ͠
࠾ج४ੜϓϩϯϓτʹର͢Δ֤Ϟσϧ͝ͱͷൺֱ ϞσϧͷܰྔԽʹΑΓ ίετͱੜ࣌ؒͷվળ Λ࣮ݱ͍ͨ͠ 1288.8×10-6 $, 5.1s 336.5×10-6 $, 3.3s ࣄલʹλΠϓͷྨ ϓϩϯϓτ࠷దԽ ʴ ίετ1/4 ੜ࣌ؒ-2s
Samba CloudΛ༻͍ͨࣄલߴਪͷ׆༻ ϦΞϧλΠϜੑΛॏࢹ͠Samba CloudΛબ ਖ਼ͷೖྗ͔Β λάΛಈతੜ ※1 gpt-oss120BͷReasoningEffortLowʹઃఆ ※2 ίϯςΩετΩϟογϡͷ༻ͳ͠
geminiͱSamba Cloudͷൺֱ 0.62s 2.9s λάΛ༻͍ͯ ࠾ج४Λੜ
ίετͱੜ࣌ؒͷվળ ࠾ج४ੜͷίετͱੜ࣌ؒ ίετɿ 40%Ҏ্ݮ ੜ࣌ؒɿ 2sվળ 742.62×10-6 $ 1288.8×10-6 $
5.1s 3.3s ※ੜ࣌ؒ࠾ج४ͷੜͷΈͰgpt-ossͷਪؚ࣌ؒΜͰ͍·ͤΜ
ಘΒΕͨݟͱࠓޙͷ։ൃ λεΫΛղܰ͠ྔͳϞσϧΛΈ߹ΘͤΔ͜ͱͰίετͱੜ͕࣌ؒվળ σʔληοτͱLLM as a judgeΛ׆༻ͨ͠Ϟσϧɾϓϩϯϓτ࠷దԽ ΑΓΠϯύΫτͷ͋Δ࠾෦ͷ׆༻(ੜె×)
͝ਗ਼ௌ͋Γ͕ͱ͏͍͟͝·ͨ͠ɻ
Appendix 1ɹGeminiͷίετ༁ Ϟσϧ input($/1Mtok) output($/1Mtok) 2.5-flash 0.3 2.5 2.5-flash-lite 0.1
0.4 Vertex AI Ͱͷ AI ϞσϧͷߏஙͱσϓϩΠʹ͔͔Δඅ༻ https://cloud.google.com/vertex-ai/generative-ai/pricing
Appendix 2ɹSamba CloudͷϞσϧൺֱɹ Ϟσϧ input($/1Mtok) output($/1Mtok) gpt-oss-120B 0.22 0.59 DeepSeek-V3.1
3.00 4.50 Qwen3-32B 0.40 0.80 Samba Cloud Pricing by Model Family & Usage Type https://cloud.sambanova.ai/plans/pricing