[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-846":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":14,"stars7d":17,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":22,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":26,"readmeContent":27,"aiSummary":28,"trendingCount":16,"starSnapshotCount":16,"syncStatus":15,"lastSyncTime":29,"discoverSource":30},846,"talkie","talkie-lm\u002Ftalkie","talkie-lm","talkie is a vintage language model from 1930","https:\u002F\u002Ftalkie-lm.com",null,"Python",900,56,4,2,0,14,101,12,74.27,"Apache License 2.0",false,"main",true,[],"2026-06-12 04:00:06","# talkie - a 13B vintage language model from 1930\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fidentity.png\" alt=\"talkie\" width=\"600\" \u002F>\n\u003C\u002Fp>\n\n`talkie` is an inference library for the talkie 13B language model family developed by Alec Radford, Nick Levine, and David Duvenaud.\n\n`talkie-1930-13b-base` is a 13b language model trained on pre-1931 English-language text.\n\n`talkie-1930-13b-it` has been instruction-tuned using a novel instruction-following dataset built from pre-1931 reference works including etiquette manuals, letter-writing manuals, encyclopedias, and poetry collections. It has also undergone reinforcement learning using online DPO to improve instruction-following capabilities. \n\nWe also provide a 'modern' base model, `talkie-web-13b-base`, with the same architecture and training FLOPs as `talkie-1930`, but trained on FineWeb, to allow for controlled comparisons between modern and vintage models. Note that we need to be careful about the claims we make contrasting the behavior and capabilities of the models, because temporal coverage is not the only difference in the pretraining corpora. For example, the distribution of subject matters differs significantly. \n\nSee our [blog post](https:\u002F\u002Ftalkie-lm.com\u002F) for details.\n\nThis package provides a simple Python API and CLI to download models from HuggingFace and run inference.\n\n## Models\n\n| Name | HuggingFace | Style | Description |\n|------|-------------|-------|-------------|\n| `talkie-1930-13b-base` | [talkie-lm\u002Ftalkie-1930-13b-base](https:\u002F\u002Fhuggingface.co\u002Ftalkie-lm\u002Ftalkie-1930-13b-base) | Base | 1930-era base language model |\n| `talkie-1930-13b-it` | [talkie-lm\u002Ftalkie-1930-13b-it](https:\u002F\u002Fhuggingface.co\u002Ftalkie-lm\u002Ftalkie-1930-13b-it) | IT | 1930-era instruction-tuned model |\n| `talkie-web-13b-base` | [talkie-lm\u002Ftalkie-web-13b-base](https:\u002F\u002Fhuggingface.co\u002Ftalkie-lm\u002Ftalkie-web-13b-base) | Base | Same architecture as talkie-1930, but trained on FineWeb |\n\n## Installation\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Ftalkie-lm\u002Ftalkie.git\ncd talkie\nuv sync\n```\n\n### Requirements\n\n- Python >= 3.11\n- PyTorch >= 2.1\n- CUDA GPU with >= 28 GB VRAM (bfloat16 inference)\n- ~26-50 GB disk space per model\n\n## Quick Start\n\n### Python API\n\n```python\nfrom talkie import Talkie\n\n# Load a base model (downloads from HuggingFace on first use)\nmodel = Talkie(\"talkie-1930-13b-base\")\n\n# Generate a completion\nresult = model.generate(\"If scientists discover life on other planets,\", temperature=0.7, max_tokens=300)\nprint(result.text)\n\n# Stream tokens\nfor token in model.stream(\"The effects of the automobile on public morality have\"):\n    print(token, end=\"\", flush=True)\n```\n\n### Chat (instruction-tuned model)\n\n```python\nfrom talkie import Talkie, Message\n\nmodel = Talkie(\"talkie-1930-13b-it\")\n\n# Single-turn\nresult = model.generate(\"Write an essay predicting what life will be like in the year 1960.\", max_tokens=600)\nprint(result.text)\n\n# Multi-turn chat\nmessages = [\n    Message(role=\"user\", content=\"What were the causes of the French Revolution?\"),\n]\nresult = model.chat(messages, temperature=0.7)\nprint(result.text)\n\n# Stream a chat reply\nmessages.append(Message(role=\"assistant\", content=result.text))\nmessages.append(Message(role=\"user\", content=\"Which of those causes was the most significant?\"))\nfor token in model.chat_stream(messages):\n    print(token, end=\"\", flush=True)\n```\n\n### Pre-download models\n\n```python\nfrom talkie import download_model\n\n# Download before loading (useful for setup scripts)\ndownload_model(\"talkie-1930-13b-base\")\n```\n\n## CLI\n\n```bash\n# Generate text\nuv run talkie generate \"Once upon a time\" --model talkie-1930-13b-base -t 0.8\n\n# Interactive chat\nuv run talkie chat --model talkie-1930-13b-it\n\n# Download a model\nuv run talkie download talkie-1930-13b-base\n\n# Download all models\nuv run talkie download all\n\n# List available models\nuv run talkie list\n```\n\n## License\n\nApache 2.0\n","talkie 是一个基于1930年代文本训练的13B参数复古语言模型。项目提供了两种主要模型：`talkie-1930-13b-base` 和 `talkie-1930-13b-it`，后者通过特定指令调优以提高遵循指令的能力，并使用在线DPO强化学习进一步优化。此外，还有一个现代版基础模型 `talkie-web-13b-base` 用于对比研究。talkie 使用Python编写，支持通过HuggingFace下载模型并进行推理。该项目适合需要模拟或研究20世纪早期语言风格和内容的场景，如历史文献分析、复古风格写作等。","2026-06-11 02:39:46","CREATED_QUERY"]