Model reference · open weights
hf-moshiko is an open-weight language model from kmhf. hf-moshiko (BF16) weighs 46.7 GB; the smallest configuration that runs it is 2× RTX 5090 32 GB.
What it is
| Released by | kmhf |
|---|---|
| Type | Language models |
| Task | Text gen |
| Parameters (lead) | 7.8B |
| Context | 3,000 tokens |
| Runs with | transformers |
| Released | 2024-09-27 |
| Popularity | 107k downloads / month |
| Weights | 46.7 GB (hf-moshiko (BF16), file size) |
| Licence | Licence not stated |
What it runs on
Weights 46.7 GB (file size) · KV cache 524 MB per 1,000 tokens of context, at 16 bits (vLLM's default for this build; an 8-bit cache halves it) · runtime overhead from 664 MB on a small card · context up to 3,000 tokens.
| Card | Requests at once 2K, its whole window tokens each | Requests at once 32K tokens each | Longest single request | Counted memory |
|---|---|---|---|---|
| RTX 3060 12 GB … L40S 48 GB 6 smaller cards | — | — | — | |
| A100 80 GB | 19 | — | all 2K | 78.2 GB |
| H100 80 GB | 17 | — | all 2K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 27 | — | all 2K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 36 | — | all 2K | 107 GB |
| H200 141 GB | 55 | — | all 2K | 138 GB |
| B200 180 GB | 79 | — | all 2K | 176 GB |
| 2× RTX 5090 32 GB tensor parallel | 8 | — | all 2K | 31.0 GB a card |
| 2× L40S 48 GB tensor parallel | 25 | — | all 2K | 44.0 GB a card |
| 4× RTX 4090 24 GB tensor parallel | 28 | — | all 2K | 23.4 GB a card |
| 4× RTX 3090 24 GB tensor parallel | 28 | — | all 2K | 23.4 GB a card |
| Requests at once | 2K, its whole window tokens each | 32K tokens each |
|---|---|---|
| 1 | 48.9 GB | — |
| 5 | 55.2 GB | — |
| 8 | 59.9 GB | — |
| 16 | 72.5 GB | — |
| 32 | 97.7 GB | — |
| 64 | 148 GB | — |
On one card, with vLLM's small-card settings (2,048 tokens a step). Cards of 70 GB and more reserve more per request and more overhead — each row above uses its own card's settings.
Estimates, not measurements, checked against published vLLM startup logs. The weights are the build's file size; the cache is calculated from its config (multi-head attention); the overhead is an estimate of vLLM's own memory with that card's default settings. "Requests at once" is how many requests of that length vLLM admits — its reservation at full length, with --max-model-len set to that length; requests that stay shorter fit more. "Longest single request" is the most one request can hold there: below the model's maximum, vLLM starts only with --max-model-len set at or under it. "Counted memory" is vLLM's default 92 % of what CUDA reports for the card (the DGX Spark: about 100 GiB of its shared 128 GB). A tensor-parallel split pools the cards' memory and speeds each token up, at the cost of the link between them; a layer split (llama.cpp) holds more but does not make one request faster. Assumes vLLM 0.10 or later.
From the model card
This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
Use the code below to get started with the model.
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
BibTeX:
[More Information Needed]
APA:
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
[More Information Needed]
Quoted from the model card on Hugging Face — the full card is behind the Hugging Face link above.