AI-Hypercomputer · copybara-service · Jul 23, 2026 · Jul 10, 2026
@@ -12,6 +12,8 @@ The following models are supported:
 | :---------------------- | :--------------------- | :-------------------: | :---------------------: | :-------------------: | :---------------------: |
 | **Gemma2**              | 2B, 9B, 27B            |           √           |            √            |           √           |            √            |
 | **Gemma3** (Multimodal) | 4B, 12B, 27B           |           √           |            √            |           √           |            √            |
+| **Gemma4**              | 26B (MoE), 31B         |           √           |            √            |           √           |            √            |
+| **Gemma4 Small**        | E2B, E4B               |           -           |            √            |           -           |            √            |
 | **Llama3.1**            | 8B, 70B, 450B          |           √           |            √            |           √           |            √            |
 | **Qwen2.5**             | 1.5B, 7B, 14B          |           √           |            √            |           √           |            √            |
 | **Qwen3**               | 0.6B, 4B, 8B, 14B, 32B |           √           |            √            |           √           |            √            |
@@ -23,6 +25,8 @@ The following models are supported:
 | **DeepSeek3.2**         | 671B                   |           √           |            √            |           -           |            -            |
 | **Qwen3 Next**          | 80B                    |           √           |            √            |           √           |            √            |
 
+> **Note:** *Gemma4 Small* (E2B / E4B) uses the `gemma4_small` decoder block, which has per-layer KV sharing and is incompatible with `nn.scan`. Only unscanned (`scan_layers=False`) conversion is supported for these variants.
+
 ## Prerequisites
 
 - MaxText must be installed in a Python virtual environment using the `maxtext[tpu]` option. For instructions on installing MaxText on your VM, please refer to the official [installation documentation](install-from-source).
@@ -37,6 +41,8 @@ Use the `to_maxtext.py` script to convert a Hugging Face model checkpoint into a
 
 > **Note:** For more information, checkout [qwen3-4b example script](https://github.com/AI-Hypercomputer/maxtext/blob/main/tests/end_to_end/tpu/qwen3/4b/test_qwen3_to_mt.sh) and [gemma3-4b example script](https://github.com/AI-Hypercomputer/maxtext/blob/main/tests/end_to_end/tpu/gemma3/4b/test_gemma3_to_mt.sh).
 
+> **Note (Gemma 4 E2B / E4B):** The `gemma4-e2b` and `gemma4-e4b` variants use the `gemma4_small` decoder block, which has per-layer KV sharing and is **incompatible with `nn.scan`**. Conversion must therefore be run with `scan_layers=False` (unscanned), and any downstream training/inference run must also use `scan_layers=False`. Additionally, if you want to convert the **base** (pretrained) model rather than the instruction-tuned default, pass `--hf_model_path=google/gemma-4-E4B` (or `google/gemma-4-E2B`) explicitly — `HF_IDS` defaults to the `-it` repository. See [convert_gemma4_base.sh](https://github.com/AI-Hypercomputer/maxtext/blob/main/tests/end_to_end/tpu/gemma4/e4b/convert_gemma4_base.sh) for a full example.
+
 ### Setup Environment
 
 ```bash
@@ -94,6 +100,8 @@ Use the `to_huggingface.py` script to convert a MaxText checkpoint into the Hugg
 
 > **Note:** For more information, checkout [qwen3-4b example script](https://github.com/AI-Hypercomputer/maxtext/blob/main/tests/end_to_end/tpu/qwen3/4b/test_qwen3_to_hf.sh).
 
+> **Note (Gemma 4 E2B / E4B):** The `gemma4-e2b` and `gemma4-e4b` variants use the `gemma4_small` decoder block and must be converted with `scan_layers=False`. If your MaxText checkpoint was fine-tuned from the base model, also pass `--hf_model_path=google/gemma-4-E4B` (or `google/gemma-4-E2B`) so the exported checkpoint bundles the base tokenizer — otherwise `to_huggingface` sources the tokenizer from `HF_IDS[gemma4-e4b]` = `google/gemma-4-E4B-it`.
+
 ### Setup Environment
 
 ```bash

@@ -0,0 +1,152 @@
+<!--
+ Copyright 2023-2026 Google LLC
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+      https://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+ -->
+
+# Reinforcement Learning with gemma4-e4b on Multi-Host TPUs
+
+This tutorial provides step-by-step instructions for setting up the environment
+and training the gemma4-e4b model with GRPO on the [OpenMathInstruct-2 dataset](https://huggingface.co/datasets/nvidia/OpenMathInstruct-2) on a Cloud TPU v6e (Trillium) GKE cluster using a `v6e-32` (4x8) slice.
+
+## Prerequisites
+
+Before starting, ensure you have:
+
+- Access to a Google Cloud Project with TPU quotas.
+- A Hugging Face account with an access token for downloading models (the `google/gemma-4-E4B` and `google/gemma-4-E4B-it` repositories are gated; request access before proceeding).
+- Permissions for Google Artifact Registry (Artifact Registry Writer role).
+- Prerequisites for XPK installed (follow [official documentation](https://github.com/AI-Hypercomputer/xpk/blob/main/docs/installation.md#1-prerequisites)).
+- A Pathways-ready GKE cluster (see [create GKE cluster](https://docs.cloud.google.com/ai-hypercomputer/docs/workloads/pathways-on-cloud/create-gke-cluster)).
+- **Docker** installed and configured for sudoless use. Follow the steps to [configure sudoless Docker](https://docs.docker.com/engine/install/linux-postinstall/).
+
+## Setup Environment Variables
+
+Set up the following environment variables to configure your training run. Replace
+placeholders with your actual values.
+
+```bash
+# Your GCP project ID.
+# If you've already set it in your local config, you can retrieve it via:
+# gcloud config get-value project
+export PROJECT_ID=<PROJECT_ID>
+
+# The name of your GKE cluster.
+export CLUSTER_NAME=<CLUSTER_NAME>
+
+# The GCP location of your GKE cluster.
+export ZONE=<ZONE> # e.g., 'us-central1' or 'us-central1-a'
+
+# Use a GCS bucket you own to store logs and checkpoints.
+export BASE_OUTPUT_DIRECTORY=<GCS_BUCKET> # e.g., gs://my-bucket/maxtext-runs
+```
+
+## Authenticate with Hugging Face
+
+To download the `gemma4-e4b` model checkpoint from Hugging Face, you need to authenticate using your Hugging Face account credentials. Run the following command and follow the prompts to log in:
+
+```bash
+hf auth login
+```
+
+## Get Your MaxText Compatible Model Checkpoint
+
+### Option 1: Using an existing MaxText checkpoint
+
+If you already have a MaxText-compatible model checkpoint, simply set the
+following environment variable and move on to the next section.
+
+```bash
+export MAXTEXT_CKPT_PATH=<CKPT_PATH> # e.g., gs://my-bucket/my-model-checkpoint/0/items
+```
+
+### Option 2: Converting from a Hugging Face checkpoint
+
+Refer to [Hugging Face to MaxText](hf-to-maxtext) to convert a Hugging Face checkpoint to MaxText format. After conversion finishes, set `MAXTEXT_CKPT_PATH` to the converted MaxText checkpoint path.
+
+```bash
+export MAXTEXT_CKPT_PATH=<CKPT_PATH> # e.g., gs://my-bucket/my-model-checkpoint/0/items
+```
+
+> **Note (Gemma 4 E4B specifics):**
+>
+> - For the `gemma4-e4b` model, you must run the conversion with `scan_layers=False` (the `gemma4_small` decoder block is incompatible with `nn.scan`); the resulting unscanned checkpoint matches the `scan_layers=False` setting used for the RL run below.
+> - This RL recipe fine-tunes the **base** model (`google/gemma-4-E4B`), not the instruction-tuned default. Pass `--hf_model_path=google/gemma-4-E4B` explicitly when running `to_maxtext` — otherwise MaxText defaults to `HF_IDS[gemma4-e4b]` = `google/gemma-4-E4B-it`.
+
+## Chat Template Configuration
+
+Unlike an instruction-tuned tokenizer, the `google/gemma-4-E4B` base tokenizer does **not** ship with a chat template, so the RL run must supply one explicitly. The [run_gemma4_e4b_rl.sh](https://github.com/AI-Hypercomputer/maxtext/blob/main/src/maxtext/trainers/post_train/rl/scripts/run_gemma4_e4b_rl.sh) script points at two files bundled with the repo (and therefore baked into your Docker image):
+
+- `data_template_path=maxtext/examples/chat_templates/openmathinstruct2_rl.json` — a stripped-down data template that injects only the system prompt and question (no turn markers). This avoids the doubled `<start_of_turn>` delimiters that would occur if the default `gsm8k_rl.json` (which bakes literal Gemma turn markers into the message content) were combined with the tokenizer's Jinja chat template.
+- `chat_template_path=maxtext/examples/chat_templates/gemma-3-27b-chat_template.json` — the Gemma 3 chat template, wrapped in a JSON object with a `chat_template` key. It is applied by the tokenizer as the single source of truth for turn-marker formatting.
+
+Both files are already included under `src/maxtext/examples/chat_templates/`, so no additional setup is required. If you want to customize the prompt formatting, edit these files (or point the config at your own) before building the Docker image.
+
+## Run RL Workload
+
+### Build and Upload MaxText Docker Image
+
+For instructions on building and uploading the MaxText Docker image with post-training dependencies, please refer to the [official documentation](../../build_maxtext.md).
+
+### Submit your workload
+
+```bash
+# The Docker image you pushed in the previous step
+export CLOUD_IMAGE_NAME=<IMAGE_NAME>
+export DOCKER_IMAGE="gcr.io/${PROJECT_ID?}/${CLOUD_IMAGE_NAME?}"
+
+# Run the RL training script on your cluster
+run_tutorial maxtext/trainers/post_train/rl/scripts/run_gemma4_e4b_rl.sh
+```
+
+> **Note:** The `run_gemma4_e4b_rl.sh` script pins the Pathways component images to specific versions via the xpk `--server-image` and `--proxy-server-image` flags (set through the `PATHWAYS_SERVER_IMAGE` and `PATHWAYS_PROXY_SERVER_IMAGE` variables at the top of the script). The `--server-image` is used for both the Pathways resource-manager server and the workers (the reference config uses the same image for both). Update these variables if you need a different Pathways release.
+
+### Monitor your workload
+
+To monitor your job's progress, you can use `kubectl` to check the `Jobset` status and stream logs directly from the pods.
+
+```bash
+kubectl get jobset -n default ${WORKLOAD_NAME}
+
+# List pods to find the specific name
+kubectl get pods | grep ${WORKLOAD_NAME}
+
+# stream the logs from the running pod (replace <POD_NAME> with the name you found)
+kubectl logs -f <POD_NAME>
+```
+
+Alternatively, after running the bash script, you will also get a link to the Google Cloud Console to view your workload logs. Follow the link to view logs and monitor your workload's progress in the Cloud Console.
+
+### Monitor RL Metrics
+
+During RL training, you can monitor key metrics to track model convergence, reward trends, and hardware performance.
+
+To enable Tunix-managed metrics measurement, set `enable_tunix_perf_metrics` to `true` in RL configurations. Note that this flag is already set to `True` by default for this tutorial workload. When enabled, Tunix automatically collects and uploads these metrics to TensorBoard.
+
+For a complete list of collected metrics, see the [Tunix Metrics Documentation](https://tunix.readthedocs.io/en/latest/metrics.html). Key metrics to monitor include:
+
+- **Model Quality & Reward Metrics:**
+  - `rewards/mean`: The average reward across the batch (crucial for tracking learning progress).
+  - `score/mean`: The average raw score from the reward model before applying the KL penalty.
+- **Rollout & Generation Metrics:**
+  - `rollout_time`: How long each rollout step takes.
+  - `completions/mean_length`: The average token length of generated completions.
+  - `actor_dequeue_time`: The time spent waiting for data from the rollout workers (relevant when async rollout is enabled).
+- **Performance & Efficiency Metrics:**
+  - `step_time_sec`: The execution time for a single training step.
+
+## Convert Checkpoint to Hugging Face Format
+
+Refer to [MaxText to Hugging Face](maxtext-to-hf) to convert a MaxText checkpoint back to Hugging Face format.
+
+> **Note (Gemma 4 E4B specifics):** Because this recipe fine-tunes the **base** model, pass `--hf_model_path=google/gemma-4-E4B` to `to_huggingface` so the exported checkpoint bundles the base tokenizer. Without it, `to_huggingface` sources the tokenizer from `HF_IDS[gemma4-e4b]` = `google/gemma-4-E4B-it` (the instruction-tuned model). Also keep `scan_layers=False`, since the `gemma4_small` decoder block is not compatible with scanned layers.
@@ -108,6 +108,15 @@
     "with values from the MaxText config. If False, raises a ValueError on mismatch.",
 )
 
+flags.DEFINE_string(
+    "hf_model_path",
+    None,
+    "(Optional) Customized remote HF repo (or local path) to source the tokenizer/processor "
+    "from. If not specified, defaults to maxtext.utils.globals.HF_IDS[model_name]. Use this to "
+    "bundle a different tokenizer than the default (e.g., the base model's tokenizer instead of "
+    "the instruction-tuned one).",
+)
+
 FLAGS = flags.FLAGS
 
 
@@ -493,7 +502,7 @@ def main(argv: Sequence[str]) -> None:
   if model_key not in HF_IDS:
     raise ValueError(f"HF Tokenizer ID not found for model key: {model_key}")
   hf_token = config.hf_access_token
-  hf_tokenizer_id = HF_IDS[model_key]
+  hf_tokenizer_id = FLAGS.hf_model_path or HF_IDS[model_key]
   tokenizer = AutoTokenizer.from_pretrained(hf_tokenizer_id, token=hf_token)
 
   # For multi-modal case:

@@ -0,0 +1,3 @@
+{
+  "chat_template": "{{ bos_token }}\n{%- if messages[0]['role'] == 'system' -%}\n    {%- if messages[0]['content'] is string -%}\n        {%- set first_user_prefix = messages[0]['content'] + '\n\n' -%}\n    {%- else -%}\n        {%- set first_user_prefix = messages[0]['content'][0]['text'] + '\n\n' -%}\n    {%- endif -%}\n    {%- set loop_messages = messages[1:] -%}\n{%- else -%}\n    {%- set first_user_prefix = \"\" -%}\n    {%- set loop_messages = messages -%}\n{%- endif -%}\n{%- for message in loop_messages -%}\n    {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}\n        {{ raise_exception(\"Conversation roles must alternate user/assistant/user/assistant/...\") }}\n    {%- endif -%}\n    {%- if (message['role'] == 'assistant') -%}\n        {%- set role = \"model\" -%}\n    {%- else -%}\n        {%- set role = message['role'] -%}\n    {%- endif -%}\n    {{ '<start_of_turn>' + role + '\n' + (first_user_prefix if loop.first else \"\") }}\n    {%- if message['content'] is string -%}\n        {{ message['content'] | trim }}\n    {%- elif message['content'] is iterable -%}\n        {%- for item in message['content'] -%}\n            {%- if item['type'] == 'image' -%}\n                {{ '<start_of_image>' }}\n            {%- elif item['type'] == 'text' -%}\n                {{ item['text'] | trim }}\n            {%- endif -%}\n        {%- endfor -%}\n    {%- else -%}\n        {{ raise_exception(\"Invalid content type\") }}\n    {%- endif -%}\n    {{ '<end_of_turn>\n' }}\n{%- endfor -%}\n{%- if add_generation_prompt -%}\n    {{'<start_of_turn>model\n'}}\n{%- endif -%}\n"
+}
@@ -0,0 +1,4 @@
+{
+  "SYSTEM_PROMPT": "You are given a problem. Think about the problem and provide your reasoning. Place it between {reasoning_start_token} and {reasoning_end_token}. Then, provide the final answer (i.e., just one numerical value) between {solution_start_token} and {solution_end_token}.",
+  "TEMPLATE": "{system_prompt}\n\n{question}"
+}