kanshan/minimind
1
0
Fork 0
mirror of https://github.com/jingyaogong/minimind.git synced 2026-05-01 11:48:14 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

[update] readme

Browse Source
This commit is contained in:
jingyaogong 2026-04-28 17:22:25 +08:00
parent 773e451b11
commit da865af63d
2 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
README.md
View File

@ -101,7 +101,7 @@
| 模型 | 参数量 | Release |
|------|--------|---------|
| minimind-3 | 64M | 2026.04.01 |
| minimind-3-moe | 198M / A64M | 2026.04.01 |
| minimind-3-moe | 198M-A64M | 2026.04.01 |
| minimind2-small | 26M | 2025.04.26 |
| minimind2-moe | 145M | 2025.04.26 |
| minimind2 | 104M | 2025.04.26 |
@ -118,7 +118,7 @@
<summary> <b>🔥 2026-04-01</b> </summary>
- 发布 `minimind-3` / `minimind-3-moe`：结构、Tokenizer、训练链路、推理接口与默认配置全面更新
- 结构主线对齐 `Qwen3 / Qwen3-MoE` 生态：Dense 约 `64M`，MoE 约 `198M / A64M`，并移除了 shared expert 设计
- 结构主线对齐 `Qwen3 / Qwen3-MoE` 生态：Dense 约 `64M`，MoE 约 `198M-A64M`，并移除了 shared expert 设计
- 默认训练数据切换为 `pretrain_t2t(_mini).jsonl`、`sft_t2t(_mini).jsonl`、`rlaif.jsonl`、`agent_rl.jsonl` 与 `agent_rl_math.jsonl`
- 移除独立 `train_reason.py`；思考能力统一由 `chat_template + <think>` 与 `open_thinking` 自适应开关控制
- `toolcall` 能力已混入 `sft_t2t / sft_t2t_mini` 主线数据，默认 `full_sft` 即具备基础 Tool Call 能力；同时新增 `scripts/chat_api.py` 等推理示例
@ -575,7 +575,7 @@ MiniMind训练数据集下载地址： [ModelScope](https://www.modelscope.cn/da
| Model Name | params | len_vocab | max_pos | rope_theta | n_layers | d_model | kv_heads | q_heads | note |
|------------|--------|-----------|---------|------------|----------|---------|----------|---------|------|
| minimind-3 | 64M | 6400 | 32768 | 1e6 | 8 | 768 | 4 | 8 | Dense |
| minimind-3-moe | 198M / A64M | 6400 | 32768 | 1e6 | 8 | 768 | 4 | 8 | 4 experts / top-1 |
| minimind-3-moe | 198M-A64M | 6400 | 32768 | 1e6 | 8 | 768 | 4 | 8 | 4 experts / top-1 |
| minimind2-small | 26M | 6400 | 32768 | 1e6 | 8 | 512 | 2 | 8 | 历史版本 |
| minimind2-moe | 145M | 6400 | 32768 | 1e6 | 8 | 640 | 2 | 8 | 历史版本 |
| minimind2 | 104M | 6400 | 32768 | 1e6 | 16 | 768 | 2 | 8 | 历史版本 |
@ -621,7 +621,7 @@ MobileLLM 的一个核心观察是：在参数量固定时，深度往往比宽
| Model Name | params | pretrain_t2t_mini | sft_t2t_mini | toolcall | RLAIF |
|------------|--------|-------------------|--------------|----------|-------|
| minimind-3 | 64M | ≈1.21h<br/>≈1.57￥ | ≈1.10h<br/>≈1.43￥ | ≈0.9h<br/>≈1.17￥ | ≈1.1h<br/>≈1.43￥ |
| minimind-3-moe | 198M / A64M | ≈1.69h<br/>≈2.20￥ | ≈1.54h<br/>≈2.00￥ | ≈1.26h<br/>≈1.64￥ | ≈1.54h<br/>≈2.00￥ |
| minimind-3-moe | 198M-A64M | ≈1.69h<br/>≈2.20￥ | ≈1.54h<br/>≈2.00￥ | ≈1.26h<br/>≈1.64￥ | ≈1.54h<br/>≈2.00￥ |
---

8
README_en.md
View File

@ -101,7 +101,7 @@ Meanwhile, third-party large model frameworks and tool libraries, such as `trans
| Model | Parameters | Release |
|------|--------|---------|
| minimind-3 | 64M | 2026.04.01 |
| minimind-3-moe | 198M / A64M | 2026.04.01 |
| minimind-3-moe | 198M-A64M | 2026.04.01 |
| minimind2-small | 26M | 2025.04.26 |
| minimind2-moe | 145M | 2025.04.26 |
| minimind2 | 104M | 2025.04.26 |
@ -117,7 +117,7 @@ Meanwhile, third-party large model frameworks and tool libraries, such as `trans
<summary> <b>🔥 2026-04-01</b> </summary>
- Released `minimind-3` / `minimind-3-moe`: comprehensive updates to structure, Tokenizer, training pipeline, inference interface, and default configuration
- Main branch structure aligned with `Qwen3 / Qwen3-MoE` ecosystem: Dense approximately `64M`, MoE approximately `198M / A64M`, and removed shared expert design
- Main branch structure aligned with `Qwen3 / Qwen3-MoE` ecosystem: Dense approximately `64M`, MoE approximately `198M-A64M`, and removed shared expert design
- Default training data switched to `pretrain_t2t(_mini).jsonl`, `sft_t2t(_mini).jsonl`, `rlaif.jsonl`, `agent_rl.jsonl`, and `agent_rl_math.jsonl`
- Removed standalone `train_reason.py`; thinking capability is now unified through `chat_template + <think>` and `open_thinking` adaptive switch control
- `toolcall` capability has been merged into `sft_t2t / sft_t2t_mini` main branch data, default `full_sft` already has basic Tool Call capability; also added inference examples such as `scripts/chat_api.py`
@ -574,7 +574,7 @@ To modify model configuration, see [./model/model_minimind.py](./model/model_min
| Model Name | params | len_vocab | max_pos | rope_theta | n_layers | d_model | kv_heads | q_heads | note |
|------------|--------|-----------|---------|------------|----------|---------|----------|---------|------|
| minimind-3 | 64M | 6400 | 32768 | 1e6 | 8 | 768 | 4 | 8 | Dense |
| minimind-3-moe | 198M / A64M | 6400 | 32768 | 1e6 | 8 | 768 | 4 | 8 | 4 experts / top-1 |
| minimind-3-moe | 198M-A64M | 6400 | 32768 | 1e6 | 8 | 768 | 4 | 8 | 4 experts / top-1 |
| minimind2-small | 26M | 6400 | 32768 | 1e6 | 8 | 512 | 2 | 8 | Historical version |
| minimind2-moe | 145M | 6400 | 32768 | 1e6 | 8 | 640 | 2 | 8 | Historical version |
| minimind2 | 104M | 6400 | 32768 | 1e6 | 16 | 768 | 2 | 8 | Historical version |
@ -620,7 +620,7 @@ For reference, GPT-3's parameter settings are as follows:
| Model Name | params | pretrain_t2t_mini | sft_t2t_mini | toolcall | RLAIF |
|------------|--------|-------------------|--------------|----------|-------|
| minimind-3 | 64M | ≈1.21h<br/>≈1.57￥ | ≈1.10h<br/>≈1.43￥ | ≈0.9h<br/>≈1.17￥ | ≈1.1h<br/>≈1.43￥ |
| minimind-3-moe | 198M / A64M | ≈1.69h<br/>≈2.20￥ | ≈1.54h<br/>≈2.00￥ | ≈1.26h<br/>≈1.64￥ | ≈1.54h<br/>≈2.00￥ |
| minimind-3-moe | 198M-A64M | ≈1.69h<br/>≈2.20￥ | ≈1.54h<br/>≈2.00￥ | ≈1.26h<br/>≈1.64￥ | ≈1.54h<br/>≈2.00￥ |
---