Trivial fixes

data/links.yml

@@ -25,7 +25,7 @@ links:
   gemma-terms:
     url: /resources/terms
   prohibited-use:
-    url: /resources/prohibited_use_policy
+    url: /resources/prohibited-use
 
   # --- Weights and collections -----------------------------------------------------------
   hf-gemma4:
@@ -122,7 +122,7 @@ links:
   gemmaverse:
     url: https://deepmind.google/models/gemma/gemmaverse/
   gemma-cookbook:
-    url: https://github.com/google-gemma/gemma-cookbook
+    url: https://github.com/google-gemma/cookbook
   gemma-skills-repo:
     url: https://github.com/google-gemma/gemma-skills
   awesome-gemma:

data/models.yml

@@ -304,15 +304,18 @@ gemma4:
         hf: google/gemma-4-E2B-it
         hf_base: google/gemma-4-E2B
         hf_qat: google/gemma-4-E2B-it-qat-w4a16-ct
-        gguf: google/gemma-4-E2B-it-qat-q4_0-gguf
+        gguf: ggml-org/gemma-4-E2B-it-GGUF
         gguf_qat: google/gemma-4-E2B-it-qat-q4_0-gguf
+        gguf_file: gemma-4-E2B-it-Q4_0.gguf
+        gguf_mmproj: mmproj-gemma-4-E2B-it-Q8_0.gguf
         ollama: gemma4:e2b
         mlx: mlx-community/gemma-4-E2B-it-qat-4bit
         lmstudio: google/gemma-4-e2b
         keras: gemma4_instruct_2b
         keras_base: gemma4_2b
         keras_assistant: gemma4_instruct_2b_assistant
         litert: litert-community/gemma-4-E2B-it-litert-lm
+        litert_file: gemma-4-E2B-it.litertlm
       memory:                          # GB, static weights + ~20% overhead
         bf16: 11.4
         sfp8: 5.7
@@ -335,15 +338,18 @@ gemma4:
         hf: google/gemma-4-E4B-it
         hf_base: google/gemma-4-E4B
         hf_qat: google/gemma-4-E4B-it-qat-w4a16-ct
-        gguf: google/gemma-4-E4B-it-qat-q4_0-gguf
+        gguf: ggml-org/gemma-4-E4B-it-GGUF
         gguf_qat: google/gemma-4-E4B-it-qat-q4_0-gguf
+        gguf_file: gemma-4-E4B-it-Q4_0.gguf
+        gguf_mmproj: mmproj-gemma-4-E4B-it-Q8_0.gguf
         ollama: gemma4:e4b
         mlx: mlx-community/gemma-4-E4B-it-qat-4bit
         lmstudio: google/gemma-4-e4b
         keras: gemma4_instruct_4b
         keras_base: gemma4_4b
         keras_assistant: gemma4_instruct_4b_assistant
         litert: litert-community/gemma-4-E4B-it-litert-lm
+        litert_file: gemma-4-E4B-it.litertlm
       memory:
         bf16: 17.9
         sfp8: 8.9
@@ -366,8 +372,10 @@ gemma4:
         hf: google/gemma-4-12B-it
         hf_base: google/gemma-4-12B
         hf_qat: google/gemma-4-12B-it-qat-w4a16-ct
-        gguf: google/gemma-4-12B-it-qat-q4_0-gguf
+        gguf: ggml-org/gemma-4-12B-it-GGUF
         gguf_qat: google/gemma-4-12B-it-qat-q4_0-gguf
+        gguf_file: gemma-4-12B-it-Q4_0.gguf
+        gguf_mmproj: mmproj-gemma-4-12B-it-Q8_0.gguf
         ollama: gemma4:12b
         mlx: mlx-community/gemma-4-12B-it-qat-4bit
         lmstudio: google/gemma-4-12b
@@ -391,8 +399,10 @@ gemma4:
       ids:                             # verified repo/tag strings; see PLAN.md
         hf: google/gemma-4-26B-A4B-it
         hf_base: google/gemma-4-26B-A4B
-        gguf: google/gemma-4-26B-A4B-it-qat-q4_0-gguf
+        gguf: ggml-org/gemma-4-26B-A4B-it-GGUF
         gguf_qat: google/gemma-4-26B-A4B-it-qat-q4_0-gguf
+        gguf_file: gemma-4-26B-A4B-it-Q4_0.gguf
+        gguf_mmproj: mmproj-gemma-4-26B-A4B-it-Q8_0.gguf
         ollama: gemma4:26b
         mlx: mlx-community/gemma-4-26B-A4B-it-qat-4bit
         lmstudio: google/gemma-4-26b-a4b
@@ -420,8 +430,10 @@ gemma4:
         hf: google/gemma-4-31B-it
         hf_base: google/gemma-4-31B
         hf_qat: google/gemma-4-31B-it-qat-w4a16-ct
-        gguf: google/gemma-4-31B-it-qat-q4_0-gguf
+        gguf: ggml-org/gemma-4-31B-it-GGUF
         gguf_qat: google/gemma-4-31B-it-qat-q4_0-gguf
+        gguf_file: gemma-4-31B-it-Q4_0.gguf
+        gguf_mmproj: mmproj-gemma-4-31B-it-Q8_0.gguf
         ollama: gemma4:31b
         mlx: mlx-community/gemma-4-31B-it-qat-4bit
         lmstudio: google/gemma-4-31b

docs/deployment/huggingface.md

@@ -12,7 +12,7 @@ colab: https://colab.research.google.com/github/google-gemma/docs/blob/main/docs
 Generating text, summarizing, and analysing content are just some of the tasks you can accomplish with Gemma open models. This tutorial shows you how to get started running Gemma using Hugging Face Transformers using both text and image input to generate text content. The Transformers Python library provides a API for accessing pre-trained generative AI models, including Gemma. For more information, see the [Transformers](https://huggingface.co/docs/transformers/en/index) documentation.
 
 !!! Note
-    Gemma 3 and later models support input of both images and text. Earlier versions of Gemma only support text input, except for some variants, including [PaliGemma](../get-started/index.md).
+    Gemma 3 and later models support input of both images and text. Earlier versions of Gemma only support text input, except for some variants, including [PaliGemma](../models/paligemma/index.md).
 
 
 ### Install Python packages
@@ -143,6 +143,7 @@ You can include multiple images in your prompt by including additional `"type":
     Do not use `<|image|>`{.token-multi}, `<start_of_image>`{.token-multi} or `<image_soft_token>`{.token-multi} tokens in text portion of a prompt template as this approach creates redundant tokens and processing errors.
 
 
+<!-- notebook:requires-start audio -->
 ## Generate text from audio data
 
 With [Gemma 4](../models/gemma-4/index.md) and [Gemma 3n](../models/gemma-3n/index.md), you can use audio data as part of your prompt. This section shows how to use the Transformers library to load and configure a Gemma model to use audio data and text input to generate text output.
@@ -151,7 +152,7 @@ With [Gemma 4](../models/gemma-4/index.md) and [Gemma 3n](../models/gemma-3n/ind
 
 When generating content with audio, use a prompt template to structure your request. A prompt template allows you to specify input from specific roles, such as `user`{.token-role} or `model`{.token-role}, and is a required format for managing multi-turn chat interactions with Gemma models. The following example code shows how to constuct a prompt template for Gemma with audio data input:
 
-```python { .notebook-cell }
+```python { .notebook-cell data-requires="audio" }
 from transformers import GenerationConfig
 config = GenerationConfig.from_pretrained(MODEL_ID)
 config.max_new_tokens = 512
@@ -183,6 +184,7 @@ You can include multiple audio files in your prompt by including additional `"ty
 !!! Note
     Do not use `<|audio|>`{.token-multi} or `<audio_soft_token>`{.token-multi} tokens in text portion of a prompt template as this approach creates redundant tokens and processing errors.
 
+<!-- notebook:requires-end -->
 
 ## Next steps
 

docs/deployment/keras.md

@@ -11,7 +11,7 @@ colab: https://colab.research.google.com/github/google-gemma/docs/blob/main/docs
 
 Generating content, summarizing, and analysing content are just some of the tasks you can accomplish with Gemma open models. This tutorial shows you how to get started running Gemma using Keras, including generating text content with text and image input. [Keras](https://keras.io/) provides implementations for running Gemma and other models using JAX, PyTorch, and TensorFlow. If you're new to Keras, you might want to read [Getting started with Keras](https://keras.io/getting_started/) before you begin.
 
-Gemma 3 and later models support text and image input. Earlier versions of Gemma only support text input, except for some variants, including [PaliGemma](../get-started/index.md).
+Gemma 3 and later models support text and image input. Earlier versions of Gemma only support text input, except for some variants, including [PaliGemma](../models/paligemma/index.md).
 !!! Note
     This notebook won't run on T4 if you try with Gemma 4. Use L4 or higher GPUs.
 

docs/deployment/llamacpp.md

@@ -13,41 +13,29 @@ most capable local runtime for Gemma 4 — **image, audio and video input all wo
 
 ## Install
 
-=== "macOS / Linux"
+=== "curl"
 
     ```bash
-    brew install llama.cpp
+    curl -LsSf https://llama.app/install.sh | sh
     ```
 
-=== "Windows"
+=== "powershell"
 
-    ```bash
-    winget install llama.cpp
-    ```
-
-=== "conda"
-
-    ```bash
-    conda install -c conda-forge llama.cpp
-    ```
-
-=== "Nix"
-
-    ```bash
-    nix profile install nixpkgs#llama-cpp
+    ```powershell
+    irm https://llama.app/install.ps1 | iex
     ```
 
 ## Run
 
 ```bash { .model-swap }
-llama-cli -hf ggml-org/gemma-4-E2B-it-GGUF --prompt "Write a poem about the Kraken."
+llama cli -hf ggml-org/gemma-4-E2B-it-GGUF --prompt "Write a poem about the Kraken."
 ```
 
 `-hf` downloads the weights on first use, including the multimodal projector. With a system
 prompt:
 
 ```bash { .model-swap }
-llama-cli -hf ggml-org/gemma-4-E2B-it-GGUF -sys "You are Hong Gildong." -p "Who are you?"
+llama cli -hf ggml-org/gemma-4-E2B-it-GGUF -sys "You are Hong Gildong." -p "Who are you?"
 ```
 
 ## Model repositories
@@ -64,21 +52,21 @@ For 4-bit, prefer Google's [QAT](../get-started/qat.md) weights — trained with
 than compressed afterwards:
 
 ```bash { .model-swap }
-llama-cli -hf google/gemma-4-12B-it-qat-q4_0-gguf -p "Explain PLE in one paragraph"
+llama cli -hf google/gemma-4-E2B-it-qat-q4_0-gguf -p "Explain gravity in one paragraph"
 ```
 
 Available as `google/gemma-4-{E2B,E4B,12B,26B-A4B,31B}-it-qat-q4_0-gguf`.
 
 ## Multimodal
 
 ```bash { .model-swap }
-llama-mtmd-cli -hf ggml-org/gemma-4-E2B-it-GGUF
+llama cli -hf ggml-org/gemma-4-E2B-it-GGUF --image "image.png" -p "Describe this image."
 ```
 
 With local files, pass the projector explicitly:
 
-```bash
-llama-server -m gemma-4-12B-it-Q4_K_M.gguf --mmproj mmproj-gemma-4-12B-it.gguf
+```bash { .model-swap }
+llama cli -m gemma-4-E2B-it-Q4_0.gguf --mm mmproj-gemma-4-E2B-it-Q8_0.gguf
 ```
 
 !!! note
@@ -88,7 +76,7 @@ llama-server -m gemma-4-12B-it-Q4_K_M.gguf --mmproj mmproj-gemma-4-12B-it.gguf
 ## Server
 
 ```bash { .model-swap }
-llama-server -hf ggml-org/gemma-4-E2B-it-GGUF
+llama server -hf ggml-org/gemma-4-E2B-it-GGUF
 ```
 
 Browser UI at `http://localhost:8080`, OpenAI-compatible API at `http://localhost:8080/v1`.

docs/deployment/unsloth.md

@@ -86,62 +86,59 @@ and both GGUF and MLX files are supported.
 
 ## llama.cpp
 
-Build `llama.cpp` first — set `-DGGML_CUDA=OFF` for CPU-only machines and for Apple silicon,
-where Metal is on by default:
+A few options to get `llama.cpp` installed on your machine:
 
-```bash
-git clone https://github.com/ggml-org/llama.cpp
-cmake llama.cpp -B llama.cpp/build \
-    -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=ON
-cmake --build llama.cpp/build --config Release -j --clean-first \
-    --target llama-cli llama-mtmd-cli llama-server llama-gguf-split
-cp llama.cpp/build/bin/llama-* llama.cpp
-```
+=== "curl"
+
+    ```bash
+    curl -LsSf https://llama.app/install.sh | sh
+    ```
+
+=== "powershell"
+
+    ```powershell
+    irm https://llama.app/install.ps1 | iex
+    ```
 
 Then pull and run a model straight from Hugging Face. There is no need to set a context length —
 llama.cpp uses exactly what is required.
 
 === "E2B"
 
     ```bash
-    export LLAMA_CACHE="unsloth/gemma-4-E2B-it-GGUF"
-    ./llama.cpp/llama-cli \
+    llama cli \
         -hf unsloth/gemma-4-E2B-it-GGUF:Q8_0 \
         --temp 1.0 --top-p 0.95 --top-k 64
     ```
 
 === "E4B"
 
     ```bash
-    export LLAMA_CACHE="unsloth/gemma-4-E4B-it-GGUF"
-    ./llama.cpp/llama-cli \
+    llama cli \
         -hf unsloth/gemma-4-E4B-it-GGUF:Q8_0 \
         --temp 1.0 --top-p 0.95 --top-k 64
     ```
 
 === "12B"
 
     ```bash
-    export LLAMA_CACHE="unsloth/gemma-4-12B-it-GGUF"
-    ./llama.cpp/llama-cli \
+    llama cli \
         -hf unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL \
         --temp 1.0 --top-p 0.95 --top-k 64
     ```
 
 === "26B A4B"
 
     ```bash
-    export LLAMA_CACHE="unsloth/gemma-4-26B-A4B-it-GGUF"
-    ./llama.cpp/llama-cli \
+    llama cli \
         -hf unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q4_K_XL \
         --temp 1.0 --top-p 0.95 --top-k 64
     ```
 
 === "31B"
 
     ```bash
-    export LLAMA_CACHE="unsloth/gemma-4-31B-it-GGUF"
-    ./llama.cpp/llama-cli \
+    llama cli \
         -hf unsloth/gemma-4-31B-it-GGUF:UD-Q4_K_XL \
         --temp 1.0 --top-p 0.95 --top-k 64
     ```
@@ -158,7 +155,7 @@ hf download unsloth/gemma-4-26B-A4B-it-GGUF \
 ```
 
 ```bash
-./llama.cpp/llama-cli \
+llama cli \
     --model unsloth/gemma-4-26B-A4B-it-GGUF/gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf \
     --mmproj unsloth/gemma-4-26B-A4B-it-GGUF/mmproj-BF16.gguf \
     --temp 1.0 --top-p 0.95 --top-k 64
@@ -167,7 +164,7 @@ hf download unsloth/gemma-4-26B-A4B-it-GGUF \
 ### Server
 
 ```bash
-./llama.cpp/llama-server \
+llama server \
     --model unsloth/gemma-4-26B-A4B-it-GGUF/gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf \
     --mmproj unsloth/gemma-4-26B-A4B-it-GGUF/mmproj-BF16.gguf \
     --temp 1.0 --top-p 0.95 --top-k 64 \

docs/get-started/capabilities/function-calling.md

@@ -264,7 +264,7 @@ if calls:
 !!! Note
     For optimal results, append the tool execution result to your message history using the specific format below. This ensures the chat template correctly generates the required token structure (e.g., `response:get_current_weather{temperature:15,weather:<|"|>sunny<|"|>}`).
 
-```python { .notebook-cell }
+```python
 "tool_responses": [
   {
     "name": function_name,
@@ -275,7 +275,7 @@ if calls:
 
 In case of multiple independent requests:
 
-```python { .notebook-cell }
+```python
 "tool_responses": [
   {
     "name": function_name_1,

docs/get-started/capabilities/image.md

@@ -320,16 +320,16 @@ output = vqa_pipe(messages, return_full_text=False, generate_kwargs=gen_kwargs)
 print(output[0]['generated_text'])
 
 draw_results(output[0]['generated_text'])
-``` 
+```
 
-```{ .no-copy .output }
+```` { .no-copy .output }
 ```json
 [
   {"box_2d": [243, 252, 956, 415], "label": "person"},
   {"box_2d": [356, 606, 654, 802], "label": "cat"}
 ]
 ```<turn|>
-```
+````
 
 ![png](../../assets/capabilities/output_b041486b_1.png)
 
@@ -409,14 +409,14 @@ print(output[0]['generated_text'])
 draw_results(output[0]['generated_text'])
 ```
 
-```{ .no-copy .output }
+```` { .no-copy .output }
 ```json
 [
   {"box_2d": [454, 871, 533, 923], "label": "person"},
   {"box_2d": [551, 316, 665, 628], "label": "car"}
 ]
 ```<turn|>
-```
+````
 
 ![png](../../assets/capabilities/output_bc2920e6_1.png)
 
@@ -506,7 +506,7 @@ plt.tight_layout()
 plt.show()
 ```
 
-```{ .no-copy .output }
+```` { .no-copy .output }
 Budget Size: 70
 # of Image Tokens: 64
 [transformers] You seem to be using the pipelines sequentially on GPU. In order to maximize efficiency please use a dataset
@@ -551,7 +551,7 @@ Budget Size: 560
   {"box_2d": [404, 867, 520, 925], "label": "person"}
 ]
 ```<turn|>
-```
+````
 
 ![png](../../assets/capabilities/output_76f17518_3.png)