Prochlo-190M-SFT is an open-weight model from CyanoAI, released under Apache License 2.0. It has 190M parameters and a 4,096-token context. At 16-bit it needs about 0.5 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
Supervised fine-tuned version of Prochlo-190M-Base — a 190.5M-parameter dense English SLM trained from scratch. All variants share architecture, tokenizer, and vocabulary.
Runs On
What it takes to serve Prochlo-190M-SFT (190M 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 | 0.4 GB | 0.5 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.2 GB | 0.2 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.1 GB | 0.1 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 18, 2026.
Prochlo-190M-SFT on every accelerator the SAVRN Index prices, at every precision
Model Card
By CyanoAI, published under apache-2.0, revision 5622815d4a9b.
Supervised fine-tuned version of Prochlo-190M-Base — a 190.5M-parameter dense English SLM trained from scratch. All variants share architecture, tokenizer, and vocabulary. 26-item behavioral rubric (format adherence, anti-sycophancy, code, concision): 61.5% pass. Strong on code instructions and factual QA. Weak on concise replies, long multi-turn coherence, and arithmetic. English only — Chinese input produces garbage by design. The chat format is plain text with control tokens, not a chat template: Note: weights are sharded across 32 safetensors files with a model.safetensors.index.json; frompretrained handles this transparently. Apache-2.0
Read CyanoAI's full model card
Supervised fine-tuned version of Prochlo-190M-Base — a 190.5M-parameter dense English SLM trained from scratch.
Model family
| Repo | Stage |
|---|---|
CyanoAI/Prochlo-190M-Base |
Pre-trained base |
CyanoAI/Prochlo-190M-SFT |
Supervised fine-tune (this repo) |
CyanoAI/Prochlo-190M |
Preference-aligned (DPO + RLAIF) — in training |
All variants share architecture, tokenizer, and vocabulary.
Training
| Base | CyanoAI/Prochlo-190M-Base (~7.7B pre-training tokens, 12 layers, vocab 50,261) |
| Data | HuggingFaceTB/smol-smoltalk, ~460k English conversations, chat-formatted with <\|user\|> / <\|assistant\|> |
| Method | Full-parameter SFT, 2 epochs, lr 3e-4 cosine, batch 216, max_len 1024, fp16, loss masked to assistant tokens |
| Steps | 96,135 |
| Final loss | 1.9308 (train), ~1.94 (eval) |
| Duration | 27 hours |
Evaluation
26-item behavioral rubric (format adherence, anti-sycophancy, code, concision): 61.5% pass.
Strong on code instructions and factual QA. Weak on concise replies, long multi-turn coherence, and arithmetic. English only — Chinese input produces garbage by design.
Usage
The chat format is plain text with control tokens, not a chat template:
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("CyanoAI/Prochlo-190M-SFT")
model = AutoModelForCausalLM.from_pretrained("CyanoAI/Prochlo-190M-SFT", dtype="float16").cuda()
prompt = "<|user|>Explain what a closure is.<|end|><|assistant|>"
ids = tok(prompt, return_tensors="pt").input_ids.cuda()
out = model.generate(ids, max_new_tokens=256, do_sample=False,
eos_token_id=tok.convert_tokens_to_ids("<|end|>"))
print(tok.decode(out[0, ids.shape[1]:], skip_special_tokens=True))
Note: weights are sharded across 32 safetensors files with a model.safetensors.index.json; from_pretrained handles this transparently.
License
Apache-2.0
Configuration
- Architecture
- Qwen3ForCausalLM
- Context length (tokens)
- 4,096
- Layers
- 12
- Hidden size
- 768
- Feed-forward size
- 3,072
- Attention heads
- 12
- Key/value heads
- 12
- Head dimension
- 64
- Vocabulary size
- 50,261
- RoPE base
- 100000
- Model type
- qwen3
Identity and Version
- Repository
- CyanoAI/Prochlo-190M-SFT
- Publisher
- CyanoAI
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- 190M parameters
- Languages
- en
- Revision
- 5622815d4a9bc5d387c4970077be04602b217471
- First published
- 2026-10-06
- Last updated
- 2026-10-07
Files and Weights
39 files, 384.5 MB in total. The weights are 32 files totalling 381.0 MB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model-00001-of-00032.safetensors | Weights | 77.2 MB | 04a61a7ed3c6 |
| model-00002-of-00032.safetensors | Weights | 77.2 MB | eea2edaa517b |
| model-00003-of-00032.safetensors | Weights | 9.4 MB | 24fe4836a4bf |
| model-00004-of-00032.safetensors | Weights | 9.4 MB | ea9b92d6c715 |
| model-00005-of-00032.safetensors | Weights | 9.4 MB | 116c9168aaa4 |
| model-00006-of-00032.safetensors | Weights | 9.4 MB | e64b059439f5 |
| model-00007-of-00032.safetensors | Weights | 9.4 MB | 4b0f97b59c90 |
| model-00008-of-00032.safetensors | Weights | 9.4 MB | 1601c70b965d |
| model-00009-of-00032.safetensors | Weights | 7.1 MB | 8b5af9fa08af |
| model-00010-of-00032.safetensors | Weights | 7.1 MB | 3deaefca6a73 |
| model-00011-of-00032.safetensors | Weights | 7.1 MB | 236a2ee2e38d |
| model-00012-of-00032.safetensors | Weights | 7.1 MB | 1f379998d787 |
| model-00013-of-00032.safetensors | Weights | 7.1 MB | cb9bbc6e1730 |
| model-00014-of-00032.safetensors | Weights | 7.1 MB | a019d0a5f324 |
| model-00015-of-00032.safetensors | Weights | 7.1 MB | ed7a59889ed4 |
| model-00016-of-00032.safetensors | Weights | 7.1 MB | 950fc7229afe |
| model-00017-of-00032.safetensors | Weights | 7.1 MB | 537629596c62 |
| model-00018-of-00032.safetensors | Weights | 7.1 MB | 354c817971c3 |
| model-00019-of-00032.safetensors | Weights | 7.1 MB | 7e66f0761cb9 |
| model-00020-of-00032.safetensors | Weights | 7.1 MB | daebae199f65 |
| model-00021-of-00032.safetensors | Weights | 7.1 MB | ce2189576e4c |
| model-00022-of-00032.safetensors | Weights | 7.1 MB | 175e1ba7b31f |
| model-00023-of-00032.safetensors | Weights | 7.1 MB | a7bd57e2251c |
| model-00024-of-00032.safetensors | Weights | 7.1 MB | 11616e77b0c5 |
| model-00025-of-00032.safetensors | Weights | 7.1 MB | 4d95125eff29 |
| model-00026-of-00032.safetensors | Weights | 7.1 MB | fbd285c44e71 |
| model-00027-of-00032.safetensors | Weights | 7.1 MB | ef496fb64089 |
| model-00028-of-00032.safetensors | Weights | 7.1 MB | 2057685c9f1b |
| model-00029-of-00032.safetensors | Weights | 7.1 MB | eda74f8196ae |
| model-00030-of-00032.safetensors | Weights | 7.1 MB | d6494bb6e55e |
| model-00031-of-00032.safetensors | Weights | 7.1 MB | d2f291f2f925 |
| model-00032-of-00032.safetensors | Weights | 7.1 MB | 7b152d3b7c14 |
| config.json | Configuration | 1.1 KB | — |
| generation_config.json | Configuration | 144 B | — |
| model.safetensors.index.json | Configuration | 10.9 KB | — |
| README.md | Documentation | 2.3 KB | — |
| .gitattributes | Repository | 1.5 KB | — |
| tokenizer.json | Tokenizer | 3.6 MB | — |
| tokenizer_config.json | Tokenizer | 664 B | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 381.0 MB
Released by CyanoAI through its official repository on Hugging Face. Read the license.
Built From
- Derived from CyanoAI/Prochlo-190M-Base
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 381.0 MB |
| 16-bit | 0.4 GB |
| 8-bit | 0.2 GB |
| 4-bit | 0.1 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About Prochlo-190M-SFT
How much GPU memory does Prochlo-190M-SFT need?
About 0.5 GB at 16-bit and 0.1 GB at 4-bit: the weights (190M parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Prochlo-190M-SFT 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.
Can I use Prochlo-190M-SFT commercially?
Yes. Prochlo-190M-SFT is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
What is Prochlo-190M-SFT's context length?
4,096 tokens, from the maximum position embeddings in its published configuration.