FITMYLLM · AUGUST 17, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
Quantization Options Select your GPU for speed estimates Quant Bits VRAM @ 8K Quality IQ2_M 2.93 27.4 GB
26.1 + 1.3 KV
low Q2_K 3.16 29.4 GB
28.1 + 1.3 KV
low IQ3_XXS 3.25 30.2 GB
28.9 + 1.3 KV
low IQ3_XS 3.5 32.4 GB
31.1 + 1.3 KV
low Q3_K_S 3.64 33.6 GB
32.3 + 1.3 KV
low IQ3_M 3.76 34.6 GB
33.4 + 1.3 KV
low Q3_K_M 4 36.7 GB
35.5 + 1.3 KV
low Q3_K_L 4.3 39.4 GB
38.1 + 1.3 KV
moderate IQ4_XS 4.46 40.8 GB
39.5 + 1.3 KV
moderate Q4_K_S 4.67 42.6 GB
41.4 + 1.3 KV
moderate Q4_K_M 4.89 44.5 GB
43.3 + 1.3 KV
good Q5_K_S 5.57 50.5 GB
49.2 + 1.3 KV
good Q5_K_M 5.7 51.6 GB
50.4 + 1.3 KV
good Q6_K 6.56 59.1 GB
57.9 + 1.3 KV
excellent Q8_0 8.5 76.1 GB
74.9 + 1.3 KV
lossless FP16 16 141.7 GB
140.5 + 1.3 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Deploying for a team or in production? Size GPUs, cost & scaling in Enterprise → ▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN DOLPHIN LLAMA 3 70B NOW
Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.
Community Ratings Chat Coding Reasoning Creative Vision Roleplay Agentic
Loading ratings...
Run this model IQ2_M — 26.1 GB VRAM Q2_K — 28.1 GB VRAM IQ3_XXS — 28.9 GB VRAM IQ3_XS — 31.1 GB VRAM Q3_K_S — 32.3 GB VRAM IQ3_M — 33.4 GB VRAM Q3_K_M — 35.5 GB VRAM Q3_K_L — 38.1 GB VRAM IQ4_XS — 39.5 GB VRAM Q4_K_S — 41.4 GB VRAM Q4_K_M — 43.3 GB VRAM Q5_K_S — 49.2 GB VRAM Q5_K_M — 50.4 GB VRAM Q6_K — 57.9 GB VRAM Q8_0 — 74.9 GB VRAM FP16 — 140.5 GB VRAM
Ollama llama.cpp vLLM LM Studio KoboldCpp Jan Docker
▸ Easiest way to get started · Beginners
DOCS ↗ curl -fsSL https://ollama.com/install.sh | shCOPY
$ ollama run dolphin-llama3:70b-v2.9-q4_K_MCOPY
Downloads and runs automatically. Add --verbose for speed stats.
▸ SETUP GUIDE >_
Auto-setup with fitmyllm CLI Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.
Auto-detect GPU Live tok/s in chat Speed benchmarks 9 inference engines
GPUs that can run this model At Q4_K_M quantization. Sorted by minimum VRAM.
Find the best GPU for Dolphin Llama 3 70B
Build Hardware for Dolphin Llama 3 70B Dolphin Llama 3 70B — uncensored 70B model. Powerful and unrestricted.
Read full model card ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
Dolphin Llama 3 70B — 70B Dense. ▸ SPECIFICATIONS
PARAMETERS 70B
ARCHITECTURE Dense Transformer
CONTEXT LENGTH 8K tokens
CAPABILITIES chat
RELEASE DATE 2024-04-20
PROVIDER Cognitive Computations
FAMILY llama ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_M 2.93 26.1 GB 75% Q2_K 3.16 28.1 GB 78% IQ3_XXS 3.25 28.9 GB 82% IQ3_XS 3.5 31.1 GB 84% Q3_K_S 3.64 32.3 GB 85% IQ3_M 3.76 33.4 GB 86% Q3_K_M 4 35.5 GB 88% Q3_K_L 4.3 38.1 GB 90% IQ4_XS 4.46 39.5 GB 92% Q4_K_S 4.67 41.4 GB 93% Q4_K_M 4.89 43.3 GB 94% Q5_K_S 5.57 49.2 GB 96% Q5_K_M 5.7 50.4 GB 96% Q6_K 6.56 57.9 GB 97% Q8_0 8.5 74.9 GB 100% FP16 16 140.5 GB 100%
§ 01 BENCHMARK SCORES
HumanEval 50.6
MMLU-PRO 48.1
MATH 48.3
IFEval 90.0
BBH 56.6
GPQA 10.5
MUSR 15.6
§ 02 RUN COMMAND
Run Dolphin Llama 3 70B locally with Ollama — needs 43.3 GB VRAM at Q4_K_M:
$ ollama run dolphin-llama3:70b
§ 03 COMPATIBLE GPUs
30 @ Q4_K_M Feedback 