Galileo Labs

User Tools

Site Tools

You are here: start » tips » llm
tips:llm

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
tips:llm [2025/12/14 17:49] – sscipionitips:llm [2025/12/26 15:35] (current) – sscipioni
Line 1: Line 1:
 ====== LLM ====== ====== LLM ======
 +
 +- https://collabnix.com/best-ollama-models-in-2025-complete-performance-comparison/
 +
 +For Production Deployment:
 +  * Primary Choice: DeepSeek-R1 32B for reasoning-heavy applications
 +  * Coding Tasks: Qwen2.5-Coder 7B for optimal balance of capability and efficiency
 +  * General Purpose: Llama 3.3 70B for maximum versatility
 +  * Edge Computing: Phi-4 14B for resource-constrained environments
 +
 +Optimization Strategies:
 +  * Always enable **Flash Attention** and KV-cache quantization
 +  * Use **Q4_K_M** quantization for production deployments
 +  * Implement caching for repeated queries
 +  * Monitor GPU memory usage and implement automatic model swapping
 +  * Use load balancing for high-throughput applications
 +
 +
 +
 +^ Hardware ^ Llama 3.3 8B (tokens/sec) ^ Llama 3.3 70B (tokens/sec) ^ Llama 3.2 ^
 +| RTX 4090 | 89.2 | 12.1 | |
 +| RTX 3090 | 67.4 | 8.3 | |
 +| A100 40GB | 156.7 | 45.2 | |
 +| M3 Max 128GB | 34.8 | 4.2 | |
 +| Strix Halo 128GB ollama | | 5.1 | 85.02 |
 +| Strix Halo 128GB llama.cpp | |  | 90 |
 +| RTX 3060 | | | 131.76 |
 +
 +
  
 ^ model                  ^ capabilities             ^ size     ^ context  ^ quantization                                                                      ^ eval rate [token/s]  ^ prompt eval rate [token/s]  ^ ^ model                  ^ capabilities             ^ size     ^ context  ^ quantization                                                                      ^ eval rate [token/s]  ^ prompt eval rate [token/s]  ^
Line 11: Line 39:
 | freehuntx/qwen3-coder:8b  | completion tools | "8.2B"   | 40960    | "Q4_K_M" | 37.97 | 565.68 | | freehuntx/qwen3-coder:8b  | completion tools | "8.2B"   | 40960    | "Q4_K_M" | 37.97 | 565.68 |
 | networkjohnny/deepseek-coder-v2-lite-base-q4_k_m-gguf:latest  | completion tools | "3.2B"   | 131072    | "Q4_K_M" | 86.02 | 1124.53 | | networkjohnny/deepseek-coder-v2-lite-base-q4_k_m-gguf:latest  | completion tools | "3.2B"   | 131072    | "Q4_K_M" | 86.02 | 1124.53 |
 +| phi4-mini  | completion tools | "3.8B"   | 131072    | "Q4_K_M" | 72.24 | 31.37 |
 +| qwen2.5:7b  | completion tools | "7.6B"   | 32768    | "Q4_K_M" | 42.98 | 153.34 |
 +| llama3.3:70b-instruct-q4_K_M  | completion tools | "70.6B"   | 131072    | "Q4_K_M" | 5.06 | 15.50 |
 +| functiongemma  | completion tools | "268.10M" | 32768 | "Q8" | 364.21 | 240.50 |
 +| danielsheep/Qwen3-Coder-30B-A3B-Instruct-1M-Unsloth  | completion tools | "30.5B"   | 1048576    | "Q4_K_M" | 71.60 | 33.14 |
 +| gpt-oss:20b  | completion tools thinking | "20.9B"   | 131072    | "MXFP4" | 47.32 | 402.47 |
  
tips/llm.1765730960.txt.gz · Last modified: by sscipioni

Page Tools

Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki