PE

perlthoughts_starling-lm-alpha-8x7b-moe

PerlthoughtsGGUF by TheBlokeUpdated 2023-12-16
Download on Hugging Face
Parameters46.7B
Context32K
Layers32
Architecturellama
Smallest quant15.6 GB
Quick answer

perlthoughts_starling-lm-alpha-8x7b-moe needs about 28.0 GB of memory at Q4_K_M (a 26.4 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 8 quants, from 15.6 GB (Q2_K) to 49.6 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.

Perlthoughts Starling LM Alpha 8X7B MoE – 46.7B parameter Llama model

This model is a 46.7B parameter Llama architecture with mixture of experts. It supports a 32768 token context window and is released under a CC‑BY‑NC‑4.0 license. Quantized versions are available from 2‑bit to 8‑bit.

Good at

  • Large context handling
  • Mixture of experts architecture
  • OpenChat‑Correct prompt template

Limits

  • Requires 8‑bit GPU for full precision
  • No vision support
  • CC‑BY‑NC‑4.0 license limits commercial use

Best for

  • Research on long‑form text generation
  • Fine‑tuning with custom datasets
  • Educational demonstrations of MoE models

Who it’s for. Researchers and developers needing a large Llama‑style model with MoE capabilities. Suitable for those who can manage the GPU memory requirements and comply with the non‑commercial license.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K15.6 GB 17.2 GB
Too big Get file →
Q3_K_M20.4 GB 21.9 GB
Too big Get file →
Q4_026.4 GB 28.0 GB
Too big Get file →
Q4_K_M26.4 GB 28.0 GB
Too big Get file →
Q5_032.2 GB 33.8 GB
Too big Get file →
Q5_K_M32.2 GB 33.8 GB
Too big Get file →
Q6_K38.4 GB 40.0 GB
Too big Get file →
Q8_049.6 GB 51.2 GB
Too big Get file →

For 8‑GB VRAM, use Q4_0 or Q4_K_M (file size 26.44 GB, memory ~28 GB). For 16‑GB VRAM, Q5_0 or Q5_K_M (32.23 GB, ~33.8 GB). For 32‑GB VRAM, Q6_K (38.38 GB, ~40 GB).

What GPU do you need for perlthoughts_starling-lm-alpha-8x7b-moe?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: no quant fits at an 8K context.
  • 6 GB VRAM: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: the best quant that fits is Q3_K_M (20.4 GB file, about 21.9 GB in use).
  • 32 GB VRAM: the best quant that fits is Q4_0 (26.4 GB file, about 28.0 GB in use).
  • 48 GB VRAM: the best quant that fits is Q6_K (38.4 GB file, about 40.0 GB in use).

Which perlthoughts_starling-lm-alpha-8x7b-moe quant should you choose?

Q4_K_M is the usual starting point for perlthoughts_starling-lm-alpha-8x7b-moe: it balances file size and quality well. Here it is a 26.4 GB file.

The Q8_0 file is 49.6 GB, about 1.9 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.

If memory is tight, the smallest file is Q2_K at 15.6 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters46.7B (approximate, mixture of experts)
Layers32
Trained context32,768 tokens
KV cache (fp16)128 KB per token

Source

Quantized byTheBloke
Downloads (HF)742
Last updated2023-12-16

Frequently asked questions about perlthoughts_starling-lm-alpha-8x7b-moe

How much VRAM does perlthoughts_starling-lm-alpha-8x7b-moe need?

About 28.0 GB at Q4_K_M with an 8K context window. That is the 26.4 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run perlthoughts_starling-lm-alpha-8x7b-moe on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 17.2 GB, so you would need a larger GPU or CPU offloading.

What is the file size of perlthoughts_starling-lm-alpha-8x7b-moe GGUF?

It depends on the quant: Q2_K is 15.6 GB, Q3_K_M is 20.4 GB, Q4_0 is 26.4 GB, Q4_K_M is 26.4 GB, Q5_0 is 32.2 GB, Q5_K_M is 32.2 GB and 2 more.

What context length does perlthoughts_starling-lm-alpha-8x7b-moe support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.

Where can I download perlthoughts_starling-lm-alpha-8x7b-moe GGUF?

The files are on Hugging Face in the TheBloke/Starling-LM-alpha-8x7B-MoE-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does perlthoughts_starling-lm-alpha-8x7b-moe use?

Hugging Face lists the license as cc-by-nc-4.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run this model on 8 GB VRAM?

Only the Q4_0 or Q4_K_M quantizations may fit, requiring ~28 GB of VRAM, so 8 GB is insufficient.

Does this model support vision tasks?

No, the model has no vision capabilities.

Is this model free for commercial use?

No, it is licensed under CC‑BY‑NC‑4.0, which prohibits commercial use.

Similar models

Browse all →

More from TheBloke

All by TheBloke →

More Perlthoughts models

All Perlthoughts →

Other large models

All large models →