Instruction-tuned checkpoint of Han2Han (han2han-ul2-base-1-it, step 43153). Han2Han is a 169M-parameter encoder-decoder model that learns script-invariant representations of Korean text: a document written in Hanja and its Hangul transcription land at the same point in embedding space. The recipe (jamo and character-level embedding fusion, morpheme-aware denoising, bidirectional Hanja-Hangul transcription) is described in the paper, accepted to Findings of EMNLP 2026. This repo holds the PyTorch weights, the SentencePiece tokenizer, and the modeling code needed to load them through the transformers Auto classes with trustremotecode=True. Training code, the Flax model, the Flax-to-PyTorch…
Open-weight model · Text generation
olmo3-190m-zh-full-continue
by Lee MikeLee000/olmo3-190m-zh-full-continue
olmo3-190m-zh-full-continue is an open-weight model for text generation from Lee, released under Apache License 2.0. It has 187M parameters and a 4,096-token context. At 16-bit it needs about 0.4 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
基于 MikeLee000/olmo3-190m-zh-full 完成一轮持续预训练（Continued Pretraining，CPT）的课程实验模型，约 187M 参数。 不是 SFT 或指令对话模型。 使用普通文本作为提示进行续写，不使用 applychattemplate，不应将其作为可靠的问答助手。 - 架构：Olmo3ForCausalLM，12 层，hidden size 768，intermediate size 3072，词表 48,000。 …
Parameters187M
Context4,096
Weights374.0 MB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads—
Runs On
What it takes to serve olmo3-190m-zh-full-continue (187M 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 | 0.4 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.
olmo3-190m-zh-full-continue on every accelerator the SAVRN Index prices, at every precision
Model Card
By Lee, published under apache-2.0, revision 5ff1914d3d7e.
基于 MikeLee000/olmo3-190m-zh-full 完成一轮持续预训练（Continued Pretraining，CPT）的课程实验模型，约 187M 参数。 不是 SFT 或指令对话模型。 使用普通文本作为提示进行续写，不使用 applychattemplate，不应将其作为可靠的问答助手。 - 架构：Olmo3ForCausalLM，12 层，hidden size 768，intermediate size 3072，词表 48,000。 - 数据：cmz1024/llm101-olmo3-zh-demo-data 仓库中的 continue/tokenizedv31.bin。 - 全参数持续预训练；不使用 LoRA；序列长度 2,048；BF16。 - 1 epoch，共 11,918 个 optimizer steps。 - Batch size 16，gradient accumulation 4，有效 batch size 64。 - Learning rate 2e-4，warmup ratio 0.1，cosine scheduler。 - Transformers 5.9.0，PyTorch 2.5.1。 - 最终 validation loss：1.9467759132385254。这是特定验证集上的语言模型损失，不是准确率或通用问答质量评分。 发布文件来自 /checkpoints/olmo3zhcontinue/final，已核对与 checkpoint-11918/model.safetensors 的 SHA-256 一致：…
Read Lee's full model card
OLMo3 Full 中文持续预训练模型
基于 MikeLee000/olmo3-190m-zh-full 完成一轮持续预训练（Continued Pretraining，CPT）的课程实验模型，约 187M 参数。
不是 SFT 或指令对话模型。 使用普通文本作为提示进行续写，不使用 apply_chat_template，不应将其作为可靠的问答助手。
训练与评估
- 架构：
Olmo3ForCausalLM，12 层，hidden size 768，intermediate size 3072，词表 48,000。 - 数据：
cmz1024/llm101-olmo3-zh-demo-data仓库中的continue/tokenized_v31.bin。 - 全参数持续预训练；不使用 LoRA；序列长度 2,048；BF16。
- 1 epoch，共 11,918 个 optimizer steps。
- Batch size 16，gradient accumulation 4，有效 batch size 64。
- Learning rate 2e-4，warmup ratio 0.1，cosine scheduler。
- Transformers 5.9.0，PyTorch 2.5.1。
- 验证集为打包序列末尾约 5%，共 40,141 段；每 1,000 步评估。
- 最终 validation loss：1.9467759132385254。这是特定验证集上的语言模型损失，不是准确率或通用问答质量评分。
权重来源
发布文件来自 /checkpoints/olmo3_zh_continue/final，已核对与 checkpoint-11918/model.safetensors 的 SHA-256 一致：
502f0a297710d58212a577a4ef20f3cfc59fc9132709cd4b3aef99d21f04576c
只发布推理所需的 safetensors 权重、配置、tokenizer 和说明，不包含训练数据、优化器状态、training_args.bin、遗留聊天模板或访问凭证。
CPU 推理示例
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "MikeLee000/olmo3-190m-zh-full-continue"
tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=False)
model = AutoModelForCausalLM.from_pretrained(
repo, dtype=torch.float32, attn_implementation="sdpa",
trust_remote_code=False,
).eval()
inputs = tokenizer("从前有座山，山里有座庙，", return_tensors="pt")
with torch.inference_mode():
output = model.generate(
**inputs, max_new_tokens=64, do_sample=True,
temperature=0.8, top_p=0.9, repetition_penalty=1.1,
use_cache=True, pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
)
print(tokenizer.decode(output[0], skip_special_tokens=True))
演示与限制
对应的 Gradio Space 使用 ZeroGPU 共享 GPU 做文本续写，不重新训练模型。GPU 按请求分配，有每日额度和排队限制；首次启动或休眠后唤醒需要等待加载。
模型可能重复、跑题、产生病句和错误事实，尤其不应相信未经核实的编程、医疗、法律或财务内容。持续预训练降低语言建模损失，不等于获得可靠的指令遵循或对话能力。请勿向公开演示输入密码、密钥或其他敏感信息。
Configuration
- Architecture
- Olmo3ForCausalLM
- Context length (tokens)
- 4,096
- Layers
- 12
- Hidden size
- 768
- Feed-forward size
- 3,072
- Attention heads
- 12
- Key/value heads
- 12
- Vocabulary size
- 48,000
- Sliding window (tokens)
- 4,096
- Model type
- olmo3
Identity and Version
- Repository
- MikeLee000/olmo3-190m-zh-full-continue
- Publisher
- Lee
- Task
- Text generation
- Modality
- Text
- Library
- transformers
- Parameters
- 187M parameters
- Languages
- zh
- Revision
- 5ff1914d3d7ef12cd7b369a01df71707725936a4
- First published
- 2026-09-30
- Last updated
- 2026-09-30
Files and Weights
8 files, 378.0 MB in total. The weights are 1 file totalling 374.0 MB in safetensors.
Weights1 file · 374.0 MB
Configuration2 files · 1.2 KB
Tokenizer2 files · 3.9 MB
Documentation1 file · 3.2 KB
Other1 file · 427 B
Repository1 file · 1.5 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 374.0 MB | 502f0a297710 |
| config.json | Configuration | 1.0 KB | — |
| generation_config.json | Configuration | 225 B | — |
| README.md | Documentation | 3.2 KB | — |
| chat_template.jinja | Other | 427 B | — |
| .gitattributes | Repository | 1.5 KB | — |
| tokenizer.json | Tokenizer | 3.9 MB | — |
| tokenizer_config.json | Tokenizer | 300 B | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 374.0 MB
Download from Lee
Released by Lee through its official repository on Hugging Face. Read the license.
Built From
- Derived from MikeLee000/olmo3-190m-zh-full
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 374.0 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 olmo3-190m-zh-full-continue
How much GPU memory does olmo3-190m-zh-full-continue need?
About 0.4 GB at 16-bit and 0.1 GB at 4-bit: the weights (187M parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run olmo3-190m-zh-full-continue 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 olmo3-190m-zh-full-continue commercially?
Yes. olmo3-190m-zh-full-continue 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 olmo3-190m-zh-full-continue's context length?
4,096 tokens, from the maximum position embeddings in its published configuration.
Similar Models
Block 0 receives the backbone's final normalized output. Blocks 1 and 2 apply their own pre-normalization. A shared, embedding-tied LM head supplies all three entropy evaluations and the final output prediction. A complete recurrent backbone, refined by three entropy-gated attention blocks. Looking for the larger model? AMOR-Mamba2 1.5B. AMOR (Adaptive Metacognitive Output Router) uses the model's predictive uncertainty to decide where attention should refine its recurrent representation. Each appended block reads normalized output entropy, compares it with a frozen threshold at inference, and adds an attention update only where its gate fires. The recurrent backbone and attention blocks…
The smallest and first model from CyanoAI — a 190.5M-parameter dense English base LM. Named after Prochlorococcus, the tiniest and most oxygen-producing photosynthetic organism on Earth: small, dense, and outproducing everything its size. All variants share the same architecture, tokenizer, and vocabulary, so they are drop-in interchangeable. Training corpus: a curated ~826GB English mixture (web / books / code / reference), stream-packed with the tokenizer jointly shipped in this repo. This is a base model — it speaks fluent Wikipedia-style prose but has not been aligned. For chat or assistant behaviour, start from the SFT or aligned variants. Zero-shot on wikitext (JSON subset, ~500k…
Block 0 receives the backbone's final normalized output. Blocks 1 and 2 apply their own pre-normalization. A shared, embedding-tied LM head supplies all three entropy evaluations and the final output prediction. A complete recurrent backbone, refined by three entropy-gated attention blocks. Looking for the larger model? AMOR-Gated DeltaNet 1.5B. AMOR (Adaptive Metacognitive Output Router) uses the model's predictive uncertainty to decide where attention should refine its recurrent representation. Each appended block reads normalized output entropy, compares it with a frozen threshold at inference, and adds an attention update only where its gate fires. The recurrent backbone and attention…
This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).