Starling LM 7B Beta — Hardware Requirements & GPU Compatibility
Chat
Starling-LM-7B-beta is Nexusflow's chat model, fine-tuned from Openchat-3.5-0106 (itself based on Mistral-7B-v0.1) using reinforcement learning from AI feedback (RLAIF). It was trained with Nexusflow's own 34B reward model and a PPO-based policy optimization pipeline on the Nectar preference dataset, and scored 8.12 on MT-Bench with GPT-4 as judge, an improvement over the team's earlier Starling model. It uses OpenChat's exact chat template rather than a generic one. At 7 billion parameters it runs comfortably on a single consumer GPU. Context length is 8,192 tokens. It is released under the Apache 2.0 license with an added condition that it not be used to compete with OpenAI, reflecting that its Nectar training data was generated with GPT-4. It was published in March 2024.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 8K)Best on 24 GB (Q4_K_M · ~133 tok/s · 8K)Macs from 8 GB (IQ3_S · ~20 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Starling LM 7B Beta
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
ollama run starling-lm:7b-beta-q4_K_MQ4_K_M fits fully on the RTX 4090 · 18.3 GB free · 8K context
Download: Starling-LM-7B-beta.Q4_K_M.gguf4.37 GBQuantFactory
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 24 GB GPU
- The easiest way to try this model
- Running in a few minutes, easy setup
- Chat right here or connect any app
$5+ bonus for new users
Runs on RTX 3080 Ti (12 GB) (6 GB headroom) · Q4_K_M
- Generation speed
- ~121 tok/s
- generation speed
- Cost per 1M output tokens
- $0.28
- per 1M output tokens
How Much VRAM Does Starling LM 7B Beta Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 3.6 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 8K context · ~180 tok/s | |
| 3.7 GB | — | 3-bit small — 3-bit small quantization | Fits: 8K context · ~175 tok/s | |
| 4.1 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 8K context · ~160 tok/s | |
| 4.2 GB | — | 4-bit · legacy — 4-bit legacy quantization | Fits: 8K context · ~156 tok/s | |
| 4.9 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 8K context · ~133 tok/s | |
| 5.7 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 8K context · ~114 tok/s | |
| 6.5 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 8K context · ~100 tok/s | |
| 7.8 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 8K context · ~84 tok/s |
Selected Q4_K_M
4-bit medium quantization5.7GBat 8K context
- Weights
- 4.3 GB
- KV cache
- 1.1 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 18.3 GB free · ~133 tok/s
≥50 tok/s needs ≥377 GB/s · RTX 5090 ~237, RTX 3090 Ti ~133
Which Hardware Can Run Starling LM 7B Beta?
Q4_K_M · 4.9 GBStarling LM 7B Beta (Q4_K_M) requires 4.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 7+ GB is recommended. Using the full 8K context window can add up to 0.8 GB, bringing total usage to 5.7 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 151 devices with unified memory can run Starling LM 7B Beta, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 2.3 GB free · up to 8K ctx
16 GPUs · 2 device families
RTX 3080~101Simulate output on RTX 3080RTX 2080 Ti~82Simulate output on RTX 2080 TiRTX 3070 Ti~81Simulate output on RTX 3070 TiOrin Nano~14Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~10Simulate output on iPhone 1720 GB
fits · 14.3 GB free · up to 8K ctx
1 GPU
RX 7900 XT~98Simulate output on RX 7900 XT32–40 GB
fits · 26.3 GB free · up to 8K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~237Simulate output on RTX 5090M5 Max36 GB~66Simulate output on Mac Studio M5 Max (36 GB)M532 GB~22Simulate output on MacBook Air 13" M5 (32 GB)Orin~27Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~18Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Starling LM 7B Beta?
llmrun Score blends public benchmark boards into one number; 100 = the average open model.
Not enough data yet for an llmrun Score.
A Score needs results on enough independent boards. How the Composite Score is built
Benchmarks
This is the single board showing results for this model; badges show who ran the test.
- LMArena Text
LMArena · Oct 2026Arena
Arena1171.0
#161/188 open
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen2.5 7B InstructAlibaba | 65.1 | 5.8 GB | ~131 | same size |
| Olmo 3 7B InstructAllen AI | — | 8.4 GB | ~114 | same size |
| Phi Mini MoE InstructMicrosoft | — | 5.4 GB | ~239 | same size |
| Falcon H1 7B InstructTII UAE | — | 6.3 GB | ~130 | same size |
| DeepSeek R1 Distill Qwen 7BDeepSeek | — | 5.8 GB | ~131 | same size |
| Mistral 7B Instruct v0.3Mistral AI | — | 6.8 GB | ~133 | same size |
Specifications
- Publisher
- Nexusflow
- Parameters
- 7.2B
- Architecture
- MistralForCausalLM
- Context Length
- 8,192 tokens
- Vocabulary Size
- 32,002
- Release Date
- 2024-03-19
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Starling LM 7B Beta need?
Starling LM 7B Beta requires 4.9 GB of VRAM at Q4_K_M, or 15.1 GB at BF16. Full 8K context adds up to 0.8 GB (5.7 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 7.2B × 4.8 bits ÷ 8 = 4.3 GB
KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 1.4 GB (at full 8K context)
VRAM usage by quantization
What's the best quantization for Starling LM 7B Beta?
For Starling LM 7B Beta, Q4_K_M (4.9 GB) offers the best balance of quality and VRAM usage. Q5_0 (5.1 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 2.6 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Starling LM 7B Beta on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Starling LM 7B Beta, the smallest at IQ3_S. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 4.9 GB of usable unified memory (RAM minus macOS overhead), from Mac Mini M2 Pro (16 GB).
Can I run Starling LM 7B Beta locally?
Yes — Starling LM 7B Beta can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Starling LM 7B Beta?
At Q4_K_M, Starling LM 7B Beta can reach ~978 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~133 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 4.91 × 0.65 = ~1059 tok/s
Estimated speed at Q4_K_M (4.9 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Starling LM 7B Beta?
At Q4_K_M, the download is about 4.35 GB. The full-precision BF16 version is 14.48 GB. The smallest option (IQ2_XXS) is 1.99 GB.
Which GPUs can run Starling LM 7B Beta?
52 consumer GPUs can run Starling LM 7B Beta at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run Starling LM 7B Beta?
151 devices with unified memory can run Starling LM 7B Beta at Q4_K_M (4.9 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.