Open-weight model
chrono4-2023-cpt-qa50-sft-forget
by Lima jjjlimaus/chrono4-2023-cpt-qa50-sft-forget
chrono4-2023-cpt-qa50-sft-forget is a model from Lima (access requested at publisher). It has 2B parameters. At 16-bit it needs about 4.8 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
2023-cutoff ChronoLLM (SN38) checkpoint trained in 3 chained stages (pretrain → sft → forget) from Every commit message is - (step counter restarts at 0 in each stage), so any revision of this repo maps to exactly one stage and step. - Optimizer carry-over.
Runs On
What it takes to serve chrono4-2023-cpt-qa50-sft-forget (2B 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 | 4.0 GB | 4.8 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 2.0 GB | 2.4 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 1.0 GB | 1.2 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 21, 2026.
chrono4-2023-cpt-qa50-sft-forget on every accelerator the SAVRN Index prices, at every precision
Model Card
2023-cutoff ChronoLLM (SN38) checkpoint trained in 3 chained stages (pretrain → sft → forget) from Every commit message is - (step counter restarts at 0 in each stage), so any revision of this repo maps to exactly one stage and step. - Optimizer carry-over. AdamW (fused) moments and fp32 master weights from the end of pretraining initialise SFT, and SFT's end state initialises forget-SFT. No optimizer reset, no re-warmup between stages. - Pretrain (CPT). Plain causal LM on 2023 news/wiki documents. Equal token quota for each month of 2023 (422,916,666 tokens/month incl. a 1.5% unread buffer), random draw inside each month, globally shuffled. Training stops at 55,486 steps. - SFT. Loss on…
Excerpt from the card by Lima.
Identity and Version
- Repository
- jjjlimaus/chrono4-2023-cpt-qa50-sft-forget
- Publisher
- Lima
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- 2B parameters
- Languages
- Not stated by the source
- Revision
- 5fdbb065c03f45de51745f51a2d68a92f98c9026
- First published
- 2026-09-30
- Last updated
- 2026-10-01
Files and Weights
7 files, 4.0 GB in total. The weights are 1 file totalling 4.0 GB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 4.0 GB | — |
| config.json | Configuration | 1.8 KB | — |
| generation_config.json | Configuration | 197 B | — |
| README.md | Documentation | 3.7 KB | — |
| .gitattributes | Repository | 1.5 KB | — |
| tokenizer.json | Tokenizer | 1.7 MB | — |
| tokenizer_config.json | Tokenizer | 196 B | — |
License and Download
- License
- Not stated by the source
- Access
- Access requested at publisher
- Download size
- 4.0 GB
Lima grants access through its official repository on Hugging Face.
Built From
- Derived from anacoluthe89/chrono4-2022
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 4.0 GB |
| 16-bit | 4.0 GB |
| 8-bit | 2.0 GB |
| 4-bit | 1.0 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Built on This Model
- Derived fromchrono4-2023-distill-mix
Questions About chrono4-2023-cpt-qa50-sft-forget
How much GPU memory does chrono4-2023-cpt-qa50-sft-forget need?
About 4.8 GB at 16-bit and 1.2 GB at 4-bit: the weights (2B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run chrono4-2023-cpt-qa50-sft-forget 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.