Model Collection
模型概览与选择建议
#TL;DR（中文）
- 这页是 的基础索引：帮助你快速建立 “哪些模型出现过、它们大致属于哪一类” 的认知地图。code
LLM - 选模型时不要只看参数/榜单：更重要的是 、cost、context length、tool support（例如code
latency）、以及你的codeTool Calling结果。codeevaluation - 在真实项目里，通常用 “capability model + fast model” 组合：复杂规划/推理用一个更强的 model，日常步骤与批处理用更快更便宜的 model。
Last updated: 2025-12
#中文导读（术语保留英文）
这份列表偏 “foundational + notable” 的历史脉络整理，不保证覆盖所有最新模型与所有版本。建议把它当作：
- 回溯用：遇到论文/博客提到某个模型时，快速定位来源与年代
- 选型用：在你做 前先建立候选池（candidate set）code
evaluation
如果你要做工程选型，建议至少回答这些问题：
- 任务更像是 、code
chat、codecoding、还是codereasoning？codeRAG - 是否需要 long context？需要多长？（context length）
- 是否需要 ？是否要 structured output（JSON schema）？code
Tool Calling - 你能否跑一个小的 集合做回归（10-50 条就够起步）？code
evaluation
Data adopted from Papers with Code and Zhao et al. (2023).
#Models
| Model | Release Date | Description |
|---|---|---|
| BERT | 2018 | Bidirectional Encoder Representations from Transformers |
| GPT | 2018 | Improving Language Understanding by Generative Pre-Training |
| RoBERTa | 2019 | A Robustly Optimized BERT Pretraining Approach |
| GPT-2 | 2019 | Language Models are Unsupervised Multitask Learners |
| T5 | 2019 | Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer |
| BART | 2019 | Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension |
| ALBERT | 2019 | A Lite BERT for Self-supervised Learning of Language Representations |
| XLNet | 2019 | Generalized Autoregressive Pretraining for Language Understanding and Generation |
| CTRL | 2019 | CTRL: A Conditional Transformer Language Model for Controllable Generation |
| ERNIE | 2019 | ERNIE: Enhanced Representation through Knowledge Integration |
| GShard | 2020 | GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding |
| GPT-3 | 2020 | Language Models are Few-Shot Learners |
| LaMDA | 2021 | LaMDA: Language Models for Dialog Applications |
| PanGu-α | 2021 | PanGu-α: Large-scale Autoregressive Pretrained Chinese Language Models with Auto-parallel Computation |
| mT5 | 2021 | mT5: A massively multilingual pre-trained text-to-text transformer |
| CPM-2 | 2021 | CPM-2: Large-scale Cost-effective Pre-trained Language Models |
| T0 | 2021 | Multitask Prompted Training Enables Zero-Shot Task Generalization |
| HyperCLOVA | 2021 | What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers |
| Codex | 2021 | Evaluating Large Language Models Trained on Code |
| ERNIE 3.0 | 2021 | ERNIE 3.0: Large-scale Knowledge Enhanced Pre-training for Language Understanding and Generation |
| Jurassic-1 | 2021 | Jurassic-1: Technical Details and Evaluation |
| FLAN | 2021 | Finetuned Language Models Are Zero-Shot Learners |
| MT-NLG | 2021 | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| Yuan 1.0 | 2021 | Yuan 1.0: Large-Scale Pre-trained Language Model in Zero-Shot and Few-Shot Learning |
| WebGPT | 2021 | WebGPT: Browser-assisted question-answering with human feedback |
| Gopher | 2021 | Scaling Language Models: Methods, Analysis & Insights from Training Gopher |
| ERNIE 3.0 Titan | 2021 | ERNIE 3.0 Titan: Exploring Larger-scale Knowledge Enhanced Pre-training for Language Understanding and Generation |
| GLaM | 2021 | GLaM: Efficient Scaling of Language Models with Mixture-of-Experts |
| InstructGPT | 2022 | Training language models to follow instructions with human feedback |
| GPT-NeoX-20B | 2022 | GPT-NeoX-20B: An Open-Source Autoregressive Language Model |
| AlphaCode | 2022 | Competition-Level Code Generation with AlphaCode |
| CodeGen | 2022 | CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis |
| Chinchilla | 2022 | Shows that for a compute budget, the best performances are not achieved by the largest models but by smaller models trained on more data. |
| Tk-Instruct | 2022 | Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks |
| UL2 | 2022 | UL2: Unifying Language Learning Paradigms |
| PaLM | 2022 | PaLM: Scaling Language Modeling with Pathways |
| OPT | 2022 | OPT: Open Pre-trained Transformer Language Models |
| BLOOM | 2022 | BLOOM: A 176B-Parameter Open-Access Multilingual Language Model |
| GLM-130B | 2022 | GLM-130B: An Open Bilingual Pre-trained Model |
| AlexaTM | 2022 | AlexaTM 20B: Few-Shot Learning Using a Large-Scale Multilingual Seq2Seq Model |
| Flan-T5 | 2022 | Scaling Instruction-Finetuned Language Models |
| Sparrow | 2022 | Improving alignment of dialogue agents via targeted human judgements |
| U-PaLM | 2022 | Transcending Scaling Laws with 0.1% Extra Compute |
| mT0 | 2022 | Crosslingual Generalization through Multitask Finetuning |
| Galactica | 2022 | Galactica: A Large Language Model for Science |
| OPT-IML | 2022 | OPT-IML: Scaling Language Model Instruction Meta Learning through the Lens of Generalization |
| LLaMA | 2023 | LLaMA: Open and Efficient Foundation Language Models |
| GPT-4 | 2023 | GPT-4 Technical Report |
| PanGu-Σ | 2023 | PanGu-Σ: Towards Trillion Parameter Language Model with Sparse Heterogeneous Computing |
| BloombergGPT | 2023 | BloombergGPT: A Large Language Model for Finance |
| PaLM 2 | 2023 | A Language Model that has better multilingual and reasoning capabilities and is more compute-efficient than its predecessor PaLM. |
