Snippet CI Workflow + Rm Quickstart section and pages

.github/workflows/check-notebookes.yaml → .github/workflows/check-notebooks.yaml

@@ -6,13 +6,17 @@ on:
       - main
     paths:
       - 'notebooks/**'
-      - '.github/workflows/check-notebookes.yaml'
+      - 'scripts/**'
+      - 'snippets/quickstart/**'
+      - '.github/workflows/check-notebooks.yaml'
   pull_request:
     branches:
       - main
     paths:
       - 'notebooks/**'
-      - '.github/workflows/check-notebookes.yaml'
+      - 'scripts/**'
+      - 'snippets/quickstart/**'
+      - '.github/workflows/check-notebooks.yaml'
   workflow_dispatch:
 
 jobs:
@@ -51,6 +55,13 @@ jobs:
           uv run nbqa mypy .
           echo "✅ Mypy type checks passed!"
 
+      - name: Check snippet freshness
+        working-directory: .
+        run: |
+          echo "🔍 Checking that generated snippets match notebook source..."
+          python3 scripts/generate_snippets.py --check
+          echo "✅ Snippets are up to date!"
+
       - name: Summary
         if: success()
         run: |

docs.json

@@ -36,7 +36,7 @@
   "logo": {
     "light": "/logo/light.svg",
     "dark": "/logo/dark.svg",
-    "href": "https://liquid.ai"
+    "href": "/docs/getting-started/welcome"
   },
   "navbar": {
     "links": [

docs/fine-tuning/trl.mdx

@@ -25,7 +25,7 @@
 
 ## Supervised Fine-Tuning (SFT)[​](#supervised-fine-tuning-sft "Direct link to Supervised Fine-Tuning (SFT)")
 
 [![Colab link](/images/lfm/fine-tuning/production/uploads/61b8e2ba285851687028d395/vlOyMEjwHa_b_LXysEu2E.png)](https://colab.research.google.com/github/Liquid4All/docs/blob/main/notebooks/💧_LFM2_5_SFT_with_TRL.ipynb)
 
 The `SFTTrainer` makes it easy to fine-tune LFM models on instruction-following or conversational datasets. It handles chat templates, packing, and dataset formatting automatically. SFT training requires [Instruction datasets](/docs/fine-tuning/datasets#instruction-datasets-sft).
 
@@ -130,18 +130,18 @@
 
 ## Vision Language Model Fine-Tuning (VLM-SFT)[​](#vision-language-model-fine-tuning-vlm-sft "Direct link to Vision Language Model Fine-Tuning (VLM-SFT)")
 
 [![Colab link](/images/lfm/fine-tuning/production/uploads/61b8e2ba285851687028d395/vlOyMEjwHa_b_LXysEu2E.png)](https://colab.research.google.com/github/Liquid4All/docs/blob/main/notebooks/💧_LFM2_5_VL_SFT_with_TRL.ipynb)
 
 The `SFTTrainer` also supports fine-tuning Vision Language Models like `LFM2.5-VL-1.6B` on image-text datasets. VLM fine-tuning requires [Vision datasets](/docs/fine-tuning/datasets#vision-datasets-vlm-sft) and a few key differences from text-only SFT:
 
 * Uses `AutoModelForImageTextToText` instead of `AutoModelForCausalLM`
 * Uses `AutoProcessor` instead of just a tokenizer
 * Requires dataset formatting with image content types
 * Needs a custom `collate_fn` for multimodal batching
 
 ### VLM LoRA Fine-Tuning (Recommended)[​](#vlm-lora-fine-tuning-recommended "Direct link to VLM LoRA Fine-Tuning (Recommended)")
 
 LoRA is recommended for VLM fine-tuning due to the larger model size and multimodal complexity:
 
 ```python
 from transformers import AutoModelForImageTextToText, AutoProcessor
@@ -288,7 +288,7 @@
 
 ## Direct Preference Optimization (DPO)[​](#direct-preference-optimization-dpo "Direct link to Direct Preference Optimization (DPO)")
 
 [![Colab link](/images/lfm/fine-tuning/production/uploads/61b8e2ba285851687028d395/vlOyMEjwHa_b_LXysEu2E.png)](https://colab.research.google.com/github/Liquid4All/docs/blob/main/notebooks/💧_LFM2_DPO_with_TRL.ipynb)
 
 The `DPOTrainer` implements Direct Preference Optimization, a method to align models with human preferences without requiring a separate reward model. DPO training requires [Preference datasets](/docs/fine-tuning/datasets#preference-datasets-dpo) with chosen and rejected response pairs.
 
@@ -385,23 +385,12 @@
   ```
 </Accordion>
 
-## Other Training Methods[​](#other-training-methods "Direct link to Other Training Methods")
-
-TRL also provides additional trainers that work seamlessly with LFM models:
-
-* **RewardTrainer**: Train reward models for RLHF
-* **PPOTrainer**: Proximal Policy Optimization for reinforcement learning from human feedback
-* **ORPOTrainer**: Odds Ratio Preference Optimization, an alternative to DPO
-* **KTOTrainer**: Kahneman-Tversky Optimization for alignment
-
-Refer to the [TRL documentation](https://huggingface.co/docs/trl) for detailed guides on these methods.
-
 ## Tips[​](#tips "Direct link to Tips")
 
 * **Learning Rates**: SFT typically uses higher learning rates (1e-5 to 5e-5) than DPO (1e-7 to 1e-6)
 * **Batch Size**: DPO requires larger effective batch sizes; increase `gradient_accumulation_steps` if GPU memory is limited
-* **LoRA Ranks**: Start with `r=16` for experimentation; increase to `r=64` or higher for better quality
-* **DPO Beta**: The `beta` parameter controls the deviation from the reference model; typical values range from 0.1 to 0.5
+* **LoRA Ranks**: Start with `r=16`. Higher ranks increase adapter memory and parameter count. Set `lora_alpha` (`a`) to `2 * r`
+* **DPO Beta**: The `beta` parameter controls the deviation from the reference model. Start with `0.1`
 
 ***
 

docs/models/complete-library.mdx

@@ -1,6 +1,6 @@
 ---
 title: "Model Library"
 description: "Liquid Foundation Models (LFMs) are a new class of multimodal architectures built for fast inference and on-device deployment. Browse all available models and formats here."
 ---
 
 <div className="capabilities">
@@ -43,15 +43,15 @@
 
 - **GGUF** — Best for local CPU/GPU inference on any platform. Use with [llama.cpp](/docs/inference/llama-cpp), [LM Studio](/docs/inference/lm-studio), or [Ollama](/docs/inference/ollama). Append `-GGUF` to any model name.
 - **MLX** — Best for Mac users with Apple Silicon. Leverages unified memory for fast inference via [MLX](/docs/inference/mlx). Browse at [mlx-community](https://huggingface.co/mlx-community/collections?search=LFM).
 - **ONNX** — Best for production deployments and edge devices. Cross-platform with ONNX Runtime across CPUs, GPUs, and accelerators. Append `-ONNX` to any model name.
 
 ### Quantization
 
 Quantization reduces model size and speeds up inference with minimal quality loss. Available options by format:
 
-- **GGUF** — Supports `Q2_K`, `Q3_K_M`, `Q4_K_M`, `Q5_K_M`, `Q6_K`, and `Q8_0` quantization levels. `Q4_K_M` offers the best balance of size and quality. `Q8_0` preserves near-full precision.
-- **MLX** — Available in `4bit` and `8bit` variants. `8bit` is the default for most models.
-- **ONNX** — Supports `FP16` and `INT8` quantization. `INT8` is best for CPU inference; `FP16` for GPU acceleration.
+- **GGUF** — Supports `Q4_0`, `Q4_K_M`, `Q5_K_M`, `Q6_K`, `Q8_0`, `BF16`, and `F16`. `Q4_K_M` offers the best balance of size and quality.
+- **MLX** — Available in `3bit`, `4bit`, `5bit`, `6bit`, `8bit`, and `BF16`. `8bit` is recommended.
+- **ONNX** — Supports `FP32`, `FP16`, `Q4`, and `Q8` (MoE models also support `Q4F16`). `Q4` is recommended for most deployments.
 
 ## Model Chart
 
@@ -82,7 +82,7 @@
 | [LFM2.5-Audio-1.5B](/docs/models/lfm25-audio-1.5b) | [✓](https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B) | [✓](https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B-GGUF) | ✗ | [✓](https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B-ONNX) | Yes (TRL) |
 | LFM2 Models | | | | | |
 | [LFM2-Audio-1.5B](/docs/models/lfm2-audio-1.5b) | [✓](https://huggingface.co/LiquidAI/LFM2-Audio-1.5B) | [✓](https://huggingface.co/LiquidAI/LFM2-Audio-1.5B-GGUF) | ✗ | ✗ | No |
 | **Liquid Nanos** | | | | | |
 | [LFM2-1.2B-Extract](/docs/models/lfm2-1.2b-extract) | [✓](https://huggingface.co/LiquidAI/LFM2-1.2B-Extract) | [✓](https://huggingface.co/LiquidAI/LFM2-1.2B-Extract-GGUF) | ✗ | [✓](https://huggingface.co/onnx-community/LFM2-1.2B-Extract-ONNX) | Yes (TRL) |
 | [LFM2-350M-Extract](/docs/models/lfm2-350m-extract) | [✓](https://huggingface.co/LiquidAI/LFM2-350M-Extract) | [✓](https://huggingface.co/LiquidAI/LFM2-350M-Extract-GGUF) | ✗ | [✓](https://huggingface.co/onnx-community/LFM2-350M-Extract-ONNX) | Yes (TRL) |
 | [LFM2-350M-ENJP-MT](/docs/models/lfm2-350m-enjp-mt) | [✓](https://huggingface.co/LiquidAI/LFM2-350M-ENJP-MT) | [✓](https://huggingface.co/LiquidAI/LFM2-350M-ENJP-MT-GGUF) | [✓](https://huggingface.co/mlx-community/LFM2-350M-ENJP-MT-8bit) | [✓](https://huggingface.co/onnx-community/LFM2-350M-ENJP-MT-ONNX) | Yes (TRL) |

notebooks/pyproject.toml

@@ -12,10 +12,6 @@ dev = [
     "mypy>=1.14.1",
 ]
 
-[tool.ruff]
-# Exclude certain directories
-extend-exclude = ["runnable-examples"]
-
 [tool.ruff.lint]
 # Select rules to check - focus on correctness, not style
 select = [

notebooks/quickstart_snippets.ipynb

@@ -0,0 +1,179 @@
+{
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "# Quickstart Snippet Sources\n",
+        "\n",
+        "This notebook is the **source of truth** for the Python code in `snippets/quickstart/*.mdx`.\n",
+        "\n",
+        "Each code cell is tagged with `\"snippet\": \"<name>\"` in its cell metadata.\n",
+        "The generation script (`scripts/generate_snippets.py`) reads these cells, replaces\n",
+        "default model names with template variables, and generates the MDX snippet files.\n",
+        "\n",
+        "**Do not edit the MDX files directly.** Edit the code cells here, then run:\n",
+        "```bash\n",
+        "python3 scripts/generate_snippets.py\n",
+        "```"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## Text Model Snippets"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "snippet": "text-transformers"
+      },
+      "outputs": [],
+      "source": [
+        "from transformers import AutoModelForCausalLM, AutoTokenizer\n",
+        "\n",
+        "model_id = \"LiquidAI/LFM2.5-1.2B-Instruct\"\n",
+        "model = AutoModelForCausalLM.from_pretrained(\n",
+        "    model_id,\n",
+        "    device_map=\"auto\",\n",
+        "    dtype=\"bfloat16\",\n",
+        ")\n",
+        "tokenizer = AutoTokenizer.from_pretrained(model_id)\n",
+        "\n",
+        "input_ids = tokenizer.apply_chat_template(\n",
+        "    [{\"role\": \"user\", \"content\": \"What is machine learning?\"}],\n",
+        "    add_generation_prompt=True,\n",
+        "    return_tensors=\"pt\",\n",
+        "    tokenize=True,\n",
+        ").to(model.device)\n",
+        "\n",
+        "output = model.generate(input_ids, max_new_tokens=512)\n",
+        "response = tokenizer.decode(output[0][len(input_ids[0]):], skip_special_tokens=True)\n",
+        "print(response)"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "snippet": "text-vllm"
+      },
+      "outputs": [],
+      "source": [
+        "from vllm import LLM, SamplingParams\n",
+        "\n",
+        "llm = LLM(model=\"LiquidAI/LFM2.5-1.2B-Instruct\")\n",
+        "\n",
+        "sampling_params = SamplingParams(\n",
+        "    temperature=0.3,\n",
+        "    min_p=0.15,\n",
+        "    repetition_penalty=1.05,\n",
+        "    max_tokens=512,\n",
+        ")\n",
+        "\n",
+        "output = llm.chat(\"What is machine learning?\", sampling_params)\n",
+        "print(output[0].outputs[0].text)"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## Vision Model Snippets"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "snippet": "vl-transformers"
+      },
+      "outputs": [],
+      "source": [
+        "from transformers import AutoProcessor, AutoModelForImageTextToText\n",
+        "from transformers.image_utils import load_image\n",
+        "\n",
+        "model_id = \"LiquidAI/LFM2.5-VL-1.6B\"\n",
+        "model = AutoModelForImageTextToText.from_pretrained(\n",
+        "    model_id,\n",
+        "    device_map=\"auto\",\n",
+        "    dtype=\"bfloat16\",\n",
+        ")\n",
+        "processor = AutoProcessor.from_pretrained(model_id)\n",
+        "\n",
+        "url = \"https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg\"\n",
+        "image = load_image(url)\n",
+        "\n",
+        "conversation = [\n",
+        "    {\n",
+        "        \"role\": \"user\",\n",
+        "        \"content\": [\n",
+        "            {\"type\": \"image\", \"image\": image},\n",
+        "            {\"type\": \"text\", \"text\": \"What is in this image?\"},\n",
+        "        ],\n",
+        "    },\n",
+        "]\n",
+        "\n",
+        "inputs = processor.apply_chat_template(\n",
+        "    conversation,\n",
+        "    add_generation_prompt=True,\n",
+        "    return_tensors=\"pt\",\n",
+        "    return_dict=True,\n",
+        "    tokenize=True,\n",
+        ").to(model.device)\n",
+        "\n",
+        "outputs = model.generate(**inputs, max_new_tokens=256)\n",
+        "response = processor.batch_decode(outputs, skip_special_tokens=True)[0]\n",
+        "print(response)"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "snippet": "vl-vllm"
+      },
+      "outputs": [],
+      "source": [
+        "from vllm import LLM, SamplingParams\n",
+        "\n",
+        "IMAGE_URL = \"http://images.cocodataset.org/val2017/000000039769.jpg\"\n",
+        "\n",
+        "llm = LLM(\n",
+        "    model=\"LiquidAI/LFM2.5-VL-1.6B\",\n",
+        "    max_model_len=1024,\n",
+        ")\n",
+        "\n",
+        "sampling_params = SamplingParams(\n",
+        "    temperature=0.0,\n",
+        "    max_tokens=256,\n",
+        ")\n",
+        "\n",
+        "messages = [{\n",
+        "    \"role\": \"user\",\n",
+        "    \"content\": [\n",
+        "        {\"type\": \"image_url\", \"image_url\": {\"url\": IMAGE_URL}},\n",
+        "        {\"type\": \"text\", \"text\": \"Describe what you see in this image.\"},\n",
+        "    ],\n",
+        "}]\n",
+        "\n",
+        "outputs = llm.chat(messages, sampling_params)\n",
+        "print(outputs[0].outputs[0].text)"
+      ]
+    }
+  ],
+  "metadata": {
+    "kernelspec": {
+      "display_name": "Python 3",
+      "name": "python3"
+    },
+    "language_info": {
+      "name": "python"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}