Hemmingway-1 27B VRAM requirements
Hemmingway-1 27B has 27.3B parameters. With an 8K-token context and one request it needs about 18.0 GB of GPU memory at Q4_K_M, 29.0 GB at FP8 and 57.0 GB at FP16/BF16. The published weights take 50.9 GB (BF16). The smallest setup here that holds it at Q4_K_M with an 8K context is the RTX 3090 (24 GB), which runs it with up to 88K tokens of context.
18.0 GB at Q4_K_M, 8K context, one request
- FP8
- 29.0 GB
- FP16 / BF16
- 57.0 GB
- Published weights
- 50.9 GB
- Smallest setup, Q4_K_M
- RTX 3090 (24 GB)
Worked example: Hemmingway-1 27B with 32K tokens
Inputs: Hemmingway-1 27B · Q4_K_M weights · 32,768 tokens of context · one request · FP16 KV cache
- Weights (27.3B at Q4_K_M)
- 15.4 GB
- KV cache (64 KB per token)
- 2.00 GB
- Buffers and runtime (0.5 GB + 10%)
- 2.24 GB
- Total
- 19.6 GB
The smallest setup here that holds it is the RTX 3090 (24 GB), with about 4.37 GB to spare. Change the inputs in the calculator
What makes Hemmingway-1 27B's memory use different
48 of its 64 layers keep a fixed-size linear-attention state, so the cache grows by 64 KB per token where caching every layer would take 256 KB.
Per token of context it adds 64 KB of FP16 cache; Granite 4.2 30B (29.3B), the nearest-sized model here with plain full attention, adds 256 KB, so Hemmingway-1 27B needs 25% as much.
How much VRAM does Hemmingway-1 27B need?
Hemmingway-1 27B needs 18.0 GB at Q4_K_M, 30.8 GB at Q8_0 and 57.0 GB at FP16 with 8,192 tokens of context and one request. Each total below is the weights plus the FP16 KV cache and the runtime overhead (0.5 GB plus 10%), and each row opens the calculator with that setting. What the GGUF names mean.
| Precision | Weights | Total | Smallest setup |
|---|---|---|---|
| As published (BF16) | 50.9 GB | 57.0 GB | 2× RTX 5090 (64 GB) |
| FP16 / BF16 | 50.9 GB | 57.0 GB | 2× RTX 5090 (64 GB) |
| FP8 / INT8 | 25.4 GB | 29.0 GB | RTX 5090 (32 GB) |
| INT4 (AWQ / GPTQ) | 13.5 GB | 15.9 GB | RTX 4060 Ti 16GB |
| GGUF Q8_0 | 27.0 GB | 30.8 GB | RTX 5090 (32 GB) |
| GGUF Q6_K | 20.9 GB | 24.0 GB | RTX 5090 (32 GB) |
| GGUF Q5_K_M | 18.0 GB | 20.9 GB | RTX 3090 (24 GB) |
| GGUF Q4_K_M | 15.4 GB | 18.0 GB | RTX 3090 (24 GB) |
| GGUF IQ4_XS | 13.8 GB | 16.3 GB | RTX 3090 (24 GB) |
| GGUF Q3_K_M | 12.4 GB | 14.7 GB | RTX 4060 Ti 16GB |
| GGUF IQ3_XXS | 10.5 GB | 12.6 GB | RTX 4060 Ti 16GB |
| GGUF Q2_K | 10.7 GB | 12.8 GB | RTX 4060 Ti 16GB |
Fine-tuning Hemmingway-1 27B? Hemmingway-1 27B VRAM for LoRA, QLoRA and full training.
Run Hemmingway-1 27B with llama.cpp
llama-server -hf bartowski/Altworld_Hemmingway-1-GGUF:Q4_K_M -c 75776 -ngl 99Altworld_Hemmingway-1-Q4_K_M.gguf, 17.4 GB, from bartowski/
Hemmingway-1 27B at 8K, 32K, 128K and 256K (full) tokens of context
16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache. Each extra token of context adds 64 KB of FP16 cache per request. The last column is the smallest setup here, one card or a group, that holds it at Q4_K_M. How the KV cache works.
| Context | KV cache, FP16 | Total at Q4_K_M | Total at Q8_0 | Smallest setup, Q4_K_M |
|---|---|---|---|---|
| 8K tokens | 512 MB | 18.0 GB | 30.8 GB | RTX 3090 (24 GB) |
| 32K tokens | 2.00 GB | 19.6 GB | 32.4 GB | RTX 3090 (24 GB) |
| 128K tokens | 8.00 GB | 26.2 GB | 39.0 GB | RTX 5090 (32 GB) |
| 256K tokens | 16.0 GB | 35.0 GB | 47.8 GB | M4 Pro Mac (64 GB, 48 GB usable) |
How fast Hemmingway-1 27B writes
Tokens per second for one request with 8,192 tokens of context, estimated from memory bandwidth. A dash means it does not fit on one card. Try other settings in the speed calculator.
| Hardware | Bandwidth | Q4_K_M | FP8 |
|---|---|---|---|
| RTX 3060 12GB | 360 GB/s | — | — |
| RTX 4090 | 1,008 GB/s | 31–43 | — |
| RTX 5090 | 1,792 GB/s | 53–76 | 34–47 |
| M4 Max Mac (128 GB) | 546 GB/s | 17–24 | 11–15 |
| M3 Ultra Mac Studio (512 GB) | 819 GB/s | 25–35 | 16–22 |
| H100 SXM | 3,350 GB/s | 93–137 | 60–86 |
| H200 | 4,800 GB/s | 126–191 | 83–121 |
Longest context on one GPU
How many tokens of context Hemmingway-1 27B fits on a single card with one request, an FP16 KV cache and 0.5 GB left free. "Full" means the model's whole context window fits.
| GPU | Memory | Q4_K_M | Q8_0 | FP8 |
|---|---|---|---|---|
| RTX 3060 | 12 GB | No | No | No |
| RTX 4060 Ti 16GB | 16 GB | No | No | No |
| RTX 3090 / 4090 | 24 GB | 88K | No | No |
| RTX 5090 | 32 GB | 204K | 18K | 43K |
| A100 40GB | 40 GB | 256K (full) | 134K | 160K |
| Mac, 64 GB unified memory | 48 GB | 256K (full) | 251K | 256K (full) |
| L40S / RTX 6000 Ada | 48 GB | 256K (full) | 251K | 256K (full) |
| A100 / H100 80GB | 80 GB | 256K (full) | 256K (full) | 256K (full) |
| Mac, 128 GB unified memory | 96 GB | 256K (full) | 256K (full) | 256K (full) |
| H200 | 141 GB | 256K (full) | 256K (full) | 256K (full) |
| B200 | 180 GB | 256K (full) | 256K (full) | 256K (full) |
Model details
- Parameters
- 27.3B (27,320,697,856)
- Layers
- 16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache
- Attention cache
- 4 KV heads × 256
- Context length
- 262,144 tokens
- Published weights
- 50.9 GB (BF16)
- On Hugging Face
- Altworld/
Hemmingway-1
Why this estimate looks this way
At Q4_K_M and an 8K-token context, Hemmingway-1 27B uses 15.4 GB for weights, 512 MB for its FP16 KV cache and 2.09 GB for estimated runtime overhead, totaling 18.0 GB. The overhead is a 0.5 GB base plus 10% of weights and cache.
Its attention layout matters for long context: 16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache. The FP16 cache grows by about 64 KB per additional token per request.
The architecture and published checkpoint size come from the model's config.json and weight files. These are estimates rather than measured peak memory; inference engines can reserve extra buffers or preallocate the full configured cache.
Hemmingway-1 27B next to similar-sized models
The three models here closest to it in parameter count, at Q4_K_M with 8,192 tokens of context.
| Model | Parameters | Total | KV per token | Smallest setup |
|---|---|---|---|---|
| Hemmingway-1 27B | 27.3B | 18.0 GB | 64 KB | RTX 3090 (24 GB) |
| Qwen3.6 27B | 27.8B | 18.3 GB | 64 KB | RTX 3090 (24 GB) |
| Qwen3.8 27B | 27.8B | 18.3 GB | 64 KB | RTX 3090 (24 GB) |
| Gemma 4 26B-A4B | 25.8B, 4B active | 17.0 GB | 20 KB | RTX 3090 (24 GB) |
Badge for your model card
Paste it into a Hugging Face or GitHub README; it shows the Q4_K_M total above and links to this page. Q8_0 and FP16 badges.
Numbers read from the model files on Hugging Face on . Compare it with other models in the reproducible model and GPU report.