Refresh Gemma 4 runtime metadata

Keep the verified W4A16 G64 RTN tensors; source Google LFS hashes are unchanged. Refresh the chat template, tokenizer metadata, processor metadata, and public model card.

README.md

@@ -1,115 +1,84 @@
 ---
 license: apache-2.0
-base_model: coder3101/gemma-4-26B-A4B-it-heretic
+license_link: https://ai.google.dev/gemma/docs/gemma_4_license
+base_model: WaveCut/gemma-4-26B-A4B-it-heretic
+base_model_relation: quantized
 library_name: vllm
+pipeline_tag: image-text-to-text
 tags:
   - gemma4
+  - heretic
   - vllm
   - llm-compressor
   - compressed-tensors
   - w4a16
-  - 4-bit
+  - int4
+  - rtn
+  - g64
   - moe
-  - rtx-3090
-pipeline_tag: text-generation
-private_benchmark: true
+  - multimodal
 ---
 
 # Gemma 4 26B A4B Heretic W4A16 G64 RTN for vLLM
 
-Private internal quantization of
-[`coder3101/gemma-4-26B-A4B-it-heretic`](https://huggingface.co/coder3101/gemma-4-26B-A4B-it-heretic)
-for vLLM on a single RTX 3090 24GB.
+`compressed-tensors` W4A16 quantization of
+[`WaveCut/gemma-4-26B-A4B-it-heretic`](https://huggingface.co/WaveCut/gemma-4-26B-A4B-it-heretic)
+for vLLM on a single 24 GB RTX 3090.
 
-This artifact was produced for AI Farm deployment experiments. It is intended
-for text-generation benchmarking and deployment validation. The vision path was
-not benchmarked.
+## Quantization
 
-## Provenance
+| Setting | Value |
+| --- | --- |
+| Method | RTN / `memoryless_minmax` |
+| Weights | signed INT4, symmetric |
+| Activations | BF16 at runtime |
+| Group size | 64 |
+| Format | `compressed-tensors`, `pack-quantized` |
+| Source architecture | `Gemma4ForConditionalGeneration` |
+| Source Heretic commit | `e6d09d5c434a5e91326ce9d779786a91e7aaae35` |
+| Original quantization stack | `llm-compressor` main, Transformers 5.8.0 |
 
-- Base model: `coder3101/gemma-4-26B-A4B-it-heretic`
-- Base commit: `e6d09d5c434a5e91326ce9d779786a91e7aaae35`
-- Source architecture: `Gemma4ForConditionalGeneration`
-- Quantization stack: `llm-compressor` main with Gemma4 MoE expert unpacking,
-  `compressed-tensors`, Transformers 5.8.0
-- Quantization method: W4A16 RTN, int4 weights, per-group, group size 64,
-  symmetric, dynamic false
-- Output format: single `model.safetensors` compressed-tensors checkpoint
-- Artifact size on disk: about 16 GiB
+Vision modules, routers, embeddings, and `lm_head` remain unquantized. Group
+size 64 is required because Gemma 4 projection widths such as 2112 and 704 are
+not divisible by 128. The exact recipe is stored in `recipe.yaml`.
 
-## Why group size 64
+## Source refresh
 
-The Gemma4 26B A4B MoE checkpoint has projection dimensions that are not
-compatible with group size 128 everywhere (`mlp.down_proj` columns 2112 and MoE
-expert down-proj columns 704). Group size 64 avoids that divisibility issue and
-was accepted by vLLM on RTX 3090 when running with fp16 activations.
+On 2026-07-16, Google updated the Gemma 4 chat template but did not update the
+model weights. The initial and current Google revisions have identical LFS
+SHA256 values for both BF16 weight shards. This repository therefore keeps the
+existing verified W4A16 tensors and refreshes the chat template, tokenizer
+metadata, and processor metadata from Google revision
+`01e5b3ee840d3a9e0b0b493c593e85398a30ef75`.
 
-## Tested vLLM Launch
+## vLLM
 
-Tested environment:
-
-- GPU: NVIDIA RTX 3090 24GB, compute capability 8.6
-- vLLM: 0.20.2
-- Torch: 2.11.0
-- CUDA runtime as seen by Torch: 13.0
+The production profile uses:
 
 ```bash
 vllm serve WaveCut/gemma-4-26B-A4B-it-heretic-W4A16-G64-RTN-vllm \
-  --served-model-name default \
-  --host 0.0.0.0 \
-  --port 8000 \
+  --served-model-name "Gemma 4 26B Heretic" \
   --trust-remote-code \
-  --dtype float16 \
+  --dtype bfloat16 \
   --max-model-len 40192 \
   --max-num-seqs 2 \
   --max-num-batched-tokens 8192 \
   --gpu-memory-utilization 0.95 \
-  --kv-cache-dtype auto
+  --generation-config vllm \
+  --no-enable-prefix-caching \
+  --enable-auto-tool-choice \
+  --tool-call-parser gemma4
 ```
 
-Notes:
-
-- `--dtype float16` is required on RTX 3090 for this WNA16 path; bf16 activation
-  selection failed kernel selection.
-- `--kv-cache-dtype fp8` failed on sm_86/Triton with unsupported `fp8e4nv`.
-- `--kv-cache-dtype fp8_e5m2` was rejected by vLLM for this compressed-tensors
-  checkpoint.
-- `--kv-cache-dtype auto` started successfully at `max_model_len=40192` and
-  `max_num_seqs=2`.
-- vLLM reported GPU KV cache size `99,951 tokens` and maximum concurrency
-  `2.49x` for `40,192` tokens/request.
-- Observed VRAM after startup was about `23.5 GiB`, so this is a tight fit on
-  RTX 3090.
-
-## Benchmark Snapshot
-
-OpenAI-compatible `/v1/completions`, two concurrent requests, `max_tokens=128`,
-temperature 0, unique prompts per request to avoid artificial prefix-cache
-reuse.
-
-| Scenario | Prompt tokens/request | Concurrency | Streaming TTFT | Streaming total | Decode after TTFT |
-| --- | ---: | ---: | ---: | ---: | ---: |
-| Empty/warm | 64 | 2 | 71-73 ms | 0.92 s | ~150.8 tok/s per request |
-| Full context | 40,000 | 2 | 11.54 s and 22.46 s | ~23.8 s | scheduler-sensitive; non-stream wall ~12.8 s |
-
-The full-context run confirms that vLLM can admit two nearly-40k requests on a
-single RTX 3090 with this artifact, but the memory margin is small and chunked
-prefill scheduling can make one request wait behind the other.
-
-## Files
-
-- `model.safetensors`: compressed W4A16 weights
-- `config.json`: Gemma4 config with compressed-tensors quantization metadata
-- `recipe.yaml`: llm-compressor recipe
-- tokenizer and chat template files
-- `preprocessor_config.json`: copied from the base model; required because vLLM
-  treats Gemma4 as a multimodal architecture during profiling even for text-only
-  serving
+The historical RTX 3090 benchmark and harness are included under `bench/` and
+`scripts/`. A fresh vLLM 0.23.0 versus 0.25.1 text, vision, and long-context
+comparison will be added after the identical A/B run completes on AI Farm.
 
 ## Limitations
 
-- RTN was used for this first working vLLM artifact; AWQ/GPTQ/other calibration
-  variants should be compared before choosing a final production quant.
-- Text-only serving was benchmarked. Image input was not validated.
-- The repository is private/internal and inherits the base model license and
-  usage constraints.
+- RTN is a weight-only quantization; compare output quality against the BF16
+  Heretic checkpoint for quality-sensitive workloads.
+- The 40,192-token, two-request profile is a tight fit on 24 GB and leaves
+  little free VRAM.
+- This is a third-party post-processed and quantized checkpoint. It is not
+  affiliated with or endorsed by Google.

chat_template.jinja

@@ -1,9 +1,15 @@
-{%- macro format_parameters(properties, required) -%}
+{#
+  Template: Google Gemma 4 Canonical Chat Template
+  Author: Google Gemma Engineering Team
+  Published: 2026-07-09
+  Context: Fixed tool-calling loops, turn closures, and thinking content-ordering.
+#}
+{%- macro format_parameters(properties, required, filter_keys=false) -%}
     {%- set standard_keys = ['description', 'type', 'properties', 'required', 'nullable'] -%}
     {%- set ns = namespace(found_first=false) -%}
     {%- for key, value in properties | dictsort -%}
         {%- set add_comma = false -%}
-        {%- if key not in standard_keys -%}
+        {%- if not filter_keys or key not in standard_keys -%}
             {%- if ns.found_first %},{% endif -%}
             {%- set ns.found_first = true -%}
             {{ key }}:{
@@ -65,7 +71,7 @@
                 {%- elif value is mapping -%}
                     {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
                     properties:{
-                    {{- format_parameters(value, value['required'] | default([])) -}}
+                    {{- format_parameters(value, value['required'] | default([]), filter_keys=true) -}}
                     }
                 {%- endif -%}
                 {%- if value['required'] -%}
@@ -116,7 +122,9 @@
     }
 {%- endmacro -%}
 {%- macro format_argument(argument, escape_keys=True) -%}
-    {%- if argument is string -%}
+    {%- if argument is none -%}
+        {{- 'null' -}}
+    {%- elif argument is string -%}
         {{- '<|"|>' + argument + '<|"|>' -}}
     {%- elif argument is boolean -%}
         {{- 'true' if argument else 'false' -}}
@@ -156,6 +164,7 @@
     {%- endfor -%}
     {{- ns.result | trim -}}
 {%- endmacro -%}
+
 {%- macro format_tool_response_block(tool_name, response) -%}
     {{- '<|tool_response>' -}}
     {%- if response is mapping -%}
@@ -170,19 +179,29 @@
     {%- endif -%}
     {{- '<tool_response|>' -}}
 {%- endmacro -%}
-{%- set ns = namespace(prev_message_type=None) -%}
+
+{#- ===== SETUP ===== -#}
+{%- set ns = namespace(prev_message_type=None, prev_non_tool_role=None) -%}
 {%- set loop_messages = messages -%}
+{%- set enable_thinking = enable_thinking | default(false) -%}
+{%- set preserve_thinking = preserve_thinking | default(false) -%}
 {{- bos_token -}}
 {#- Handle System/Tool Definitions Block -#}
-{%- if (enable_thinking is defined and enable_thinking) or tools or messages[0]['role'] in ['system', 'developer'] -%}
+{%- if enable_thinking or tools or (messages and messages[0]['role'] in ['system', 'developer']) -%}
     {{- '<|turn>system\n' -}}
     {#- Inject Thinking token at the very top of the FIRST system turn -#}
-    {%- if enable_thinking is defined and enable_thinking -%}
+    {%- if enable_thinking -%}
         {{- '<|think|>\n' -}}
         {%- set ns.prev_message_type = 'think' -%}
     {%- endif -%}
-    {%- if messages[0]['role'] in ['system', 'developer'] -%}
-        {{- messages[0]['content'] | trim -}}
+    {%- if messages and messages[0]['role'] in ['system', 'developer'] -%}
+        {%- if messages[0]['content'] is string -%}
+            {{- messages[0]['content'] | trim -}}
+        {%- elif messages[0]['content'] is sequence -%}
+            {%- for item in messages[0]['content'] -%}
+                {{- item['text'] | trim + ' '-}}
+            {%- endfor -%}
+        {%- endif -%}
         {%- set loop_messages = messages[1:] -%}
     {%- endif -%}
     {%- if tools -%}
@@ -195,6 +214,7 @@
     {%- endif -%}
     {{- '<turn|>\n' -}}
 {%- endif %}
+
 {#- Pre-scan: find last user message index for reasoning guard -#}
 {%- set ns_turn = namespace(last_user_idx=-1) -%}
 {%- for i in range(loop_messages | length) -%}
@@ -202,34 +222,28 @@
         {%- set ns_turn.last_user_idx = i -%}
     {%- endif -%}
 {%- endfor -%}
+
 {#- Loop through messages -#}
 {%- for message in loop_messages -%}
     {%- if message['role'] != 'tool' -%}
     {%- set ns.prev_message_type = None -%}
     {%- set role = 'model' if message['role'] == 'assistant' else message['role'] -%}
-    {#- Detect continuation: suppress duplicate <|turn>model when previous non-tool message was also assistant -#}
-    {%- set prev_nt = namespace(role=None, found=false) -%}
-    {%- if loop.index0 > 0 -%}
-        {%- for j in range(loop.index0 - 1, -1, -1) -%}
-            {%- if not prev_nt.found -%}
-                {%- if loop_messages[j]['role'] != 'tool' -%}
-                    {%- set prev_nt.role = loop_messages[j]['role'] -%}
-                    {%- set prev_nt.found = true -%}
-                {%- endif -%}
-            {%- endif -%}
-        {%- endfor -%}
-    {%- endif -%}
-    {%- set continue_same_model_turn = (role == 'model' and prev_nt.role == 'assistant') -%}
+    {#- Detect continuation using tracked state — O(1) instead of O(n) backward scan -#}
+    {%- set continue_same_model_turn = (role == 'model' and ns.prev_non_tool_role == 'assistant') -%}
     {%- if not continue_same_model_turn -%}
         {{- '<|turn>' + role + '\n' }}
+
     {%- endif -%}
+
     {#- Render reasoning/reasoning_content as thinking channel -#}
     {%- set thinking_text = message.get('reasoning') or message.get('reasoning_content') -%}
-    {%- if thinking_text and loop.index0 > ns_turn.last_user_idx and message.get('tool_calls') -%}
+    {%- set thinking_gate = (loop.index0 > ns_turn.last_user_idx) or (preserve_thinking and message.get('tool_calls')) -%}
+    {%- if thinking_text and thinking_gate -%}
         {{- '<|channel>thought\n' + thinking_text + '\n<channel|>' -}}
     {%- endif -%}
-            {%- if message['tool_calls'] -%}
-                {%- for tool_call in message['tool_calls'] -%}
+
+            {%- if message.get('tool_calls') -%}
+                {%- for tool_call in message.get('tool_calls') -%}
                     {%- set function = tool_call['function'] -%}
                     {{- '<|tool_call>call:' + function['name'] + '{' -}}
                     {%- if function['arguments'] is mapping -%}
@@ -239,18 +253,24 @@
                             {%- set ns_args.found_first = true -%}
                             {{- key -}}:{{- format_argument(value, escape_keys=False) -}}
                         {%- endfor -%}
-                    {%- elif function['arguments'] is string -%}
-                        {{- function['arguments'] -}}
+                    {%- elif function['arguments'] is none -%}
+                    {%- else -%}
+                        {{- raise_exception(
+                            "chat_template: tool_calls[].function.arguments must be a "
+                            "JSON object (mapping), not a string. Deserialize arguments "
+                            "before passing to the template."
+                        ) -}}
                     {%- endif -%}
                     {{- '}<tool_call|>' -}}
                 {%- endfor -%}
                 {%- set ns.prev_message_type = 'tool_call' -%}
             {%- endif -%}
+
             {%- set ns_tr_out = namespace(flag=false) -%}
             {%- if message.get('tool_responses') -%}
                 {#- Legacy: tool_responses embedded on the assistant message (Google/Gemma native) -#}
-                {%- for tool_response in message['tool_responses'] -%}
-                    {{- format_tool_response_block(tool_response['name'] | default('unknown'), tool_response['response']) -}}
+                {%- for tool_response in message.get('tool_responses') -%}
+                    {{- format_tool_response_block(tool_response['name'] | default('unknown', true), tool_response['response']) -}}
                     {%- set ns_tr_out.flag = true -%}
                     {%- set ns.prev_message_type = 'tool_response' -%}
                 {%- endfor -%}
@@ -264,8 +284,8 @@
                     {%- else -%}
                         {%- set follow = loop_messages[k] -%}
                         {#- Resolve tool_call_id to function name -#}
-                        {%- set ns_tname = namespace(name=follow.get('name') | default('unknown')) -%}
-                        {%- for tc in message['tool_calls'] -%}
+                        {%- set ns_tname = namespace(name=follow.get('name') or 'unknown') -%}
+                        {%- for tc in message.get('tool_calls') -%}
                             {%- if tc.get('id') == follow.get('tool_call_id') -%}
                                 {%- set ns_tname.name = tc['function']['name'] -%}
                             {%- endif -%}
@@ -282,6 +302,15 @@
                                 {%- endif -%}
                             {%- endfor -%}
                             {{- format_tool_response_block(ns_tname.name, ns_txt.s) -}}
+                            {%- for part in tool_body -%}
+                                {%- if part.get('type') in ['image', 'image_url'] -%}
+                                    {{- '<|image|>' -}}
+                                {%- elif part.get('type') in ['audio', 'input_audio'] -%}
+                                    {{- '<|audio|>' -}}
+                                {%- elif part.get('type') == 'video' -%}
+                                    {{- '<|video|>' -}}
+                                {%- endif -%}
+                            {%- endfor -%}
                         {%- else -%}
                             {{- format_tool_response_block(ns_tname.name, tool_body) -}}
                         {%- endif -%}
@@ -290,44 +319,72 @@
                     {%- endif -%}
                 {%- endfor -%}
             {%- endif -%}
-            {%- if message['content'] is string -%}
+
+            {%- set captured_content -%}
+            {%- if message.get('content') is string -%}
                 {%- if role == 'model' -%}
                     {{- strip_thinking(message['content']) -}}
                 {%- else -%}
                     {{- message['content'] | trim -}}
                 {%- endif -%}
-            {%- elif message['content'] is sequence -%}
+            {%- elif message.get('content') is sequence -%}
                 {%- for item in message['content'] -%}
-                    {%- if item['type'] == 'text' -%}
+                    {%- if item.get('type') == 'text' -%}
                         {%- if role == 'model' -%}
                             {{- strip_thinking(item['text']) -}}
                         {%- else -%}
                             {{- item['text'] | trim -}}
                         {%- endif -%}
-                    {%- elif item['type'] == 'image' -%}
+                    {%- elif item.get('type') in ['image', 'image_url'] -%}
                         {{- '<|image|>' -}}
-                        {%- set ns.prev_message_type = 'image' -%}
-                    {%- elif item['type'] == 'audio' -%}
+                    {%- elif item.get('type') in ['audio', 'input_audio'] -%}
                         {{- '<|audio|>' -}}
-                        {%- set ns.prev_message_type = 'audio' -%}
-                    {%- elif item['type'] == 'video' -%}
+                    {%- elif item.get('type') == 'video' -%}
                         {{- '<|video|>' -}}
-                        {%- set ns.prev_message_type = 'video' -%}
                     {%- endif -%}
                 {%- endfor -%}
             {%- endif -%}
+            {%- endset -%}
+
+            {{- captured_content -}}
+            {%- set has_content = captured_content | trim | length > 0 -%}
+
+        {#- Forward-scan: find next non-tool message role for continuation detection -#}
+        {%- set next_nt = namespace(role=None, found=false) -%}
+        {%- for j in range(loop.index0 + 1, loop_messages | length) -%}
+            {%- if not next_nt.found -%}
+                {%- if loop_messages[j]['role'] != 'tool' -%}
+                    {%- set next_nt.role = loop_messages[j]['role'] -%}
+                    {%- set next_nt.found = true -%}
+                {%- endif -%}
+            {%- endif -%}
+        {%- endfor -%}
+
+        {%- set continues_into_next = (
+            role == 'model'
+            and next_nt.role == 'assistant'
+            and (not message.get('tool_calls') or ns_tr_out.flag)
+        ) -%}
+
         {%- if ns.prev_message_type == 'tool_call' and not ns_tr_out.flag -%}
             {{- '<|tool_response>' -}}
-        {%- elif not (ns_tr_out.flag and not message.get('content')) -%}
+        {%- elif continues_into_next -%}
+        {%- elif not (ns_tr_out.flag and not has_content and not next_nt.found) -%}
             {{- '<turn|>\n' -}}
         {%- endif -%}
+
+    {#- Track previous non-tool role for next iteration (avoids O(n) backward scan) -#}
+    {%- set ns.prev_non_tool_role = message['role'] -%}
     {%- endif -%}
 {%- endfor -%}
+
 {%- if add_generation_prompt -%}
     {%- if ns.prev_message_type != 'tool_response' and ns.prev_message_type != 'tool_call' -%}
         {{- '<|turn>model\n' -}}
-        {%- if not enable_thinking | default(false) -%}
+        {%- if not enable_thinking -%}
             {{- '<|channel>thought\n<channel|>' -}}
         {%- endif -%}
+    {%- elif ns.prev_message_type == 'tool_response' and enable_thinking -%}
+        {{- '<|channel>thought\n' -}}
     {%- endif -%}
-{%- endif -%}
+{%- endif -%}

processor_config.json

@@ -0,0 +1,75 @@
+{
+  "audio_ms_per_token": 40,
+  "audio_seq_length": 750,
+  "feature_extractor": {
+    "dither": 0.0,
+    "feature_extractor_type": "Gemma4AudioFeatureExtractor",
+    "feature_size": 128,
+    "fft_length": 512,
+    "fft_overdrive": false,
+    "frame_length": 320,
+    "hop_length": 160,
+    "input_scale_factor": 1.0,
+    "max_frequency": 8000.0,
+    "mel_floor": 0.001,
+    "min_frequency": 0.0,
+    "padding_side": "right",
+    "padding_value": 0.0,
+    "per_bin_mean": null,
+    "per_bin_stddev": null,
+    "preemphasis": 0.0,
+    "preemphasis_htk_flavor": true,
+    "return_attention_mask": true,
+    "sampling_rate": 16000
+  },
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": false,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.0,
+      0.0,
+      0.0
+    ],
+    "image_processor_type": "Gemma4ImageProcessor",
+    "image_seq_length": 280,
+    "image_std": [
+      1.0,
+      1.0,
+      1.0
+    ],
+    "max_soft_tokens": 280,
+    "patch_size": 16,
+    "pooling_kernel_size": 3,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098
+  },
+  "image_seq_length": 280,
+  "processor_class": "Gemma4Processor",
+  "video_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "image_mean": [
+      0.0,
+      0.0,
+      0.0
+    ],
+    "image_std": [
+      1.0,
+      1.0,
+      1.0
+    ],
+    "max_soft_tokens": 70,
+    "num_frames": 32,
+    "patch_size": 16,
+    "pooling_kernel_size": 3,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "video_processor_type": "Gemma4VideoProcessor"
+  }
+}

tokenizer_config.json

@@ -17,72 +17,50 @@
     "<|video|>"
   ],
   "image_token": "<|image|>",
-  "is_local": true,
-  "local_files_only": false,
   "mask_token": "<mask>",
   "model_max_length": 1000000000000000019884624838656,
-  "model_specific_special_tokens": {
-    "audio_token": "<|audio|>",
-    "boa_token": "<|audio>",
-    "boi_token": "<|image>",
-    "eoa_token": "<audio|>",
-    "eoc_token": "<channel|>",
-    "eoi_token": "<image|>",
-    "eot_token": "<turn|>",
-    "escape_token": "<|\"|>",
-    "etc_token": "<tool_call|>",
-    "etd_token": "<tool|>",
-    "etr_token": "<tool_response|>",
-    "image_token": "<|image|>",
-    "soc_token": "<|channel>",
-    "sot_token": "<|turn>",
-    "stc_token": "<|tool_call>",
-    "std_token": "<|tool>",
-    "str_token": "<|tool_response>",
-    "think_token": "<|think|>"
-  },
   "pad_token": "<pad>",
   "padding_side": "left",
   "processor_class": "Gemma4Processor",
   "response_schema": {
+    "type": "object",
     "properties": {
-      "content": {
-        "type": "string"
-      },
       "role": {
         "const": "assistant"
       },
       "thinking": {
         "type": "string"
       },
+      "content": {
+        "type": "string"
+      },
       "tool_calls": {
+        "x-regex-iterator": "<\\|tool_call>(.*?)<tool_call\\|>",
+        "type": "array",
         "items": {
+          "type": "object",
           "properties": {
+            "type": {
+              "const": "function"
+            },
             "function": {
+              "type": "object",
+              "x-regex": "call\\:(?P<name>\\w+)(?P<arguments>\\{.*\\})",
               "properties": {
-                "arguments": {
-                  "additionalProperties": {},
-                  "type": "object",
-                  "x-parser": "gemma4-tool-call"
-                },
                 "name": {
                   "type": "string"
+                },
+                "arguments": {
+                  "type": "object",
+                  "x-parser": "gemma4-tool-call",
+                  "additionalProperties": {}
                 }
-              },
-              "type": "object",
-              "x-regex": "call\\:(?P<name>\\w+)(?P<arguments>\\{.*\\})"
-            },
-            "type": {
-              "const": "function"
+              }
             }
-          },
-          "type": "object"
-        },
-        "type": "array",
-        "x-regex-iterator": "<\\|tool_call>(.*?)<tool_call\\|>"
+          }
+        }
       }
     },
-    "type": "object",
     "x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<tool_calls>\\<\\|tool_call\\>.*\\<tool_call\\|\\>)?(?P<content>(?:(?!\\<turn\\|\\>)(?!\\<\\|tool_response\\>).)+)?(?:\\<turn\\|\\>|\\<\\|tool_response\\>)?"
   },
   "soc_token": "<|channel>",