Open-weight model
llama31-8b-16x4-sni-emo_random_pool-s42
by Jongbin Won Jongbin-kr/llama31-8b-16x4-sni-emo_random_pool-s42
llama31-8b-16x4-sni-emo_random_pool-s42 is an open-weight model from Jongbin Won. It has 8.5B parameters and a 131,072-token context. At 16-bit it needs about 20.4 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
SNI EMO random-pool seed 42. This repository contains standalone Transformers model snapshots at its root across ordered Git commits. Epochs 1–5 were committed in order.
Runs On
What it takes to serve llama31-8b-16x4-sni-emo_random_pool-s42 (8.5B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 17.0 GB | 20.4 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 8.5 GB | 10.2 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 4.2 GB | 5.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 19, 2026.
Model Card
SNI EMO random-pool seed 42. This repository contains standalone Transformers model snapshots at its root across ordered Git commits. Epochs 1–5 were committed in order. The final main revision and best tag point to the minimum validation completion CE checkpoint, epoch 2. Load the current best model with AutoModelForCausalLM.frompretrained(repoid, trustremotecode=True, torchdtype='auto', devicemap='auto'). Set revision='epoch-N' to load a specific epoch. Training sampled a pool size uniformly from 3 through 15 per example; checkpoint selection and inference use the full 15-expert pool. The model has 15 routed experts, top-3 routed selection, one always-active shared expert, and LoRA rank…
Excerpt from the card by Jongbin Won.
Configuration
- Architecture
- LlamaEmoRoutedLoraForCausalLM
- Context length (tokens)
- 131,072
- Layers
- 32
- Hidden size
- 4,096
- Feed-forward size
- 14,336
- Attention heads
- 32
- Key/value heads
- 8
- Head dimension
- 128
- Vocabulary size
- 128,256
- Experts active per token
- 4
- Model type
- llama_emo_routed_lora
Identity and Version
- Repository
- Jongbin-kr/llama31-8b-16x4-sni-emo_random_pool-s42
- Publisher
- Jongbin Won
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- 8.5B parameters
- Languages
- sni, moe
- Revision
- c3bfa2ea89050d3de81d81f0f8160400dbf218b1
- First published
- 2026-10-16
- Last updated
- 2026-10-16
Files and Weights
17 files, 17.0 GB in total. The weights are 5 files totalling 17.0 GB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model-00001-of-00005.safetensors | Weights | 4.0 GB | dc107251b344 |
| model-00002-of-00005.safetensors | Weights | 4.0 GB | 5a77a8a7e20a |
| model-00003-of-00005.safetensors | Weights | 3.9 GB | d8c8b0ae75fa |
| model-00004-of-00005.safetensors | Weights | 4.0 GB | f045342ed4dc |
| model-00005-of-00005.safetensors | Weights | 1.1 GB | e2b5de021319 |
| checkpoint_history.json | Configuration | 633 B | — |
| checkpoint_metadata.json | Configuration | 1.2 KB | — |
| config.json | Configuration | 1.5 KB | — |
| configuration_llama_emo_routed_lora.py | Configuration | 3.5 KB | — |
| generation_config.json | Configuration | 183 B | — |
| model.safetensors.index.json | Configuration | 41.6 KB | — |
| modeling_llama_emo_routed_lora.py | Configuration | 16.7 KB | — |
| README.md | Documentation | 1.5 KB | — |
| chat_template.jinja | Other | 4.6 KB | — |
| .gitattributes | Repository | 1.6 KB | — |
| tokenizer.json | Tokenizer | 17.2 MB | 6b9e4e7fb171 |
| tokenizer_config.json | Tokenizer | 323 B | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 17.0 GB
Released by Jongbin Won through its official repository on Hugging Face.
Built From
- Derived from Jongbin-kr/llama3.1-8b-16x4-emo-style-routed-lora-moe
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 17.0 GB |
| 16-bit | 17.0 GB |
| 8-bit | 8.5 GB |
| 4-bit | 4.2 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About llama31-8b-16x4-sni-emo_random_pool-s42
How much GPU memory does llama31-8b-16x4-sni-emo_random_pool-s42 need?
About 20.4 GB at 16-bit and 5.1 GB at 4-bit: the weights (8.5B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run llama31-8b-16x4-sni-emo_random_pool-s42 on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
What is llama31-8b-16x4-sni-emo_random_pool-s42's context length?
131,072 tokens, from the maximum position embeddings in its published configuration.