Publish AX-Qwen3-Coder-Next-MLX-AXQ-MXFP4 factory convert

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+axquant_plan.json filter=lfs diff=lfs merge=lfs -text
+axquant_quantizer_execution.json filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,184 @@
+---
+license: apache-2.0
+library_name: mlx
+base_model: Qwen/Qwen3-Coder-Next
+base_model_relation: quantized
+pipeline_tag: text-generation
+tags:
+- mlx
+- apple-silicon
+- quantized
+- mixed-precision
+- axquant
+- axq
+- development
+- qwen3-next
+- MXFP4
+- MXFP4
+---
+
+# AX-Qwen3-Coder-Next-MLX-AXQ-MXFP4 — 4.32 BPW measured main
+
+An **AXQuant (AXQ)** mixed-precision MLX checkpoint for Apple Silicon, converted directly from
+the BF16 source model. The language path is quantized under AXQuant protection floors (embeddings, norms, and other protected tensors remain higher precision).
+
+> **Checkpoint Tier 1 certified** on `df-macstudio-m2` (2026-08-18) for this exact
+> revision — measured size against a matched uniform baseline, quality retention, and
+> conversion integrity. Tier 1 is a checkpoint claim, **not** a speed claim: MTP
+> acceleration is **not certified**; no MTP speedup claim for this checkpoint.
+> See the [checkpoint Tier 1 certificate](https://github.com/defai-digital/axquant/blob/main/docs/certifications/qwen3-coder-next-axq-mxfp4-tier1.md) for the bound evidence and thresholds.
+
+
+## Model details
+
+| Property | Value |
+| --- | --- |
+| Base model | [Qwen/Qwen3-Coder-Next](https://huggingface.co/Qwen/Qwen3-Coder-Next/tree/a7fbcb5c0e12d62a448eaa0e260346bf5dcc0feb) |
+| Source revision | `a7fbcb5c0e12d62a448eaa0e260346bf5dcc0feb` |
+| Product family | `qwen3-next` |
+| Source architecture | `Qwen3NextForCausalLM` (mixture of experts (MoE)); text path optimized |
+| Main-model parameters | 79.67B logical parameters |
+| Quantizer | AXQuant `1.9.0` |
+| Hub budget class | `MXFP4` |
+| AXQuant base precision class | `5p6bpw` |
+| Planned storage-adjusted BPW | 5.0590 |
+| Measured main-model BPW | 4.3154 |
+| Measured total BPW | **4.3154** |
+| Safetensors weight size | 42.98 GB |
+| Approximate complete download | 43.08 GB |
+| Configured maximum context | 262,144 tokens; practical limits depend on unified memory |
+| Primary MLX runtime | MLX-LM |
+| AX Engine native execution | Not established; no validated native manifest is included |
+| MTP present | `False` |
+| Vision present | `False` |
+| Audio present | `False` |
+
+This repository contains MLX Safetensors. It does **not** contain PyTorch or GGUF weights.
+
+## Choosing an AXQ pack
+
+AXQ names describe a **storage-budget product class**, not one uniform precision applied to every
+tensor. Protected tensors remain at higher precision, so the exact measured BPW is authoritative.
+In particular, a `6bit`-named mixed plan may retain `4bit` as its base precision while selecting
+6-bit, 8-bit, or BF16 for other tensors to meet an approximately 6-BPW total budget. Protection
+floors can also raise a `4bit`-named pack close to (or above) a `6bit` budget on small or heavily
+protected models. When that collapse happens, AutomatosX does **not** publish a separate
+misleading `4bit` sibling for that base.
+
+
+| Sibling | Intended trade-off |
+| --- | --- |
+| [4bit sibling](https://huggingface.co/AutomatosX/AX-Qwen3-Coder-Next-MLX-AXQ-4bit) | Lower-storage AXQ budget; check its exact BPW |
+| [6bit sibling](https://huggingface.co/AutomatosX/AX-Qwen3-Coder-Next-MLX-AXQ-6bit) | Higher average precision near the 6-BPW budget |
+
+See the [AutomatosX collections](https://huggingface.co/AutomatosX/collections)
+for the family catalog, or the [complete index](https://huggingface.co/collections/AutomatosX/automatosx-mlx-model-catalog).
+
+## Download
+
+```bash
+python -m pip install -U huggingface_hub
+hf download AutomatosX/AX-Qwen3-Coder-Next-MLX-AXQ-MXFP4 --local-dir ./AX-Qwen3-Coder-Next-MLX-AXQ-MXFP4
+```
+
+Allow at least 43.08 GB of free disk space. Pin the resulting Hub commit in reproducible
+deployments rather than relying indefinitely on `main`.
+
+## Run with MLX-LM
+
+```bash
+python -m pip install -U mlx-lm
+mlx_lm.generate \
+  --model AutomatosX/AX-Qwen3-Coder-Next-MLX-AXQ-MXFP4 \
+  --prompt "Explain mixed-precision quantization in three sentences." \
+  --max-tokens 128 \
+  --temp 0.0
+```
+
+MLX-LM compatibility covers standard **text/backbone inference**. It may ignore AXQuant runtime
+metadata and optional sidecars (`vision.safetensors`, `mtp.safetensors`); this command therefore
+does not establish MTP acceleration or vision-language quality. The artifact records MLX
+`0.32.0` and MLX-LM `0.31.3` from conversion.
+
+## AX Engine status
+
+This package does **not** include a validated native `model-manifest.json`, so AX Engine execution
+is not established by this release. The AX Engine fields in `axquant_runtime.json` describe the
+intended compatibility contract, not observed runtime evidence. Use the architecture-specific MLX
+runtime path above. The artifact records AX Engine version
+`not recorded`, but version discovery alone is not a runtime check.
+
+## Quantization layout
+
+| Main-weight precision | Parameters | Share |
+| --- | ---: | ---: |
+| `4bit` | 79.00B | 99.15% |
+| `8bit` | 361.59M | 0.45% |
+| `bf16` | 312.56M | 0.39% |
+
+- Quantization methods: `affine, bf16`.
+- Group sizes used by quantized assignments: `32, 64`.
+- MTP sidecar: not included.
+- Vision sidecar: not included.
+- Optimization scope: `text-path`.
+- Support tier: `convertible`.
+
+BF16 sidecars, when present, are included in total download size. Their presence does not by itself
+establish MTP acceleration or vision-language quality.
+
+## Evidence and validation status
+
+| Check | Status |
+| --- | --- |
+| Planning evidence | `architecture_prior` |
+| Calibration | none; the allocation is based on architecture priors |
+| Quantizer execution | 74125/74125 recorded module conversions succeeded; 0 fallbacks |
+| AX Engine native manifest | not included |
+| Quality versus BF16 or uniform baselines | Not published; no quality-retention claim |
+| MTP acceptance and speed | **not certified**; no MTP speedup claim for this checkpoint (Qwen3-Coder-Next source declares no MTP; certification is non-MTP direct-decode checkpoint Tier 1 only (qwen3-next-direct track).) |
+| AX Engine kernel evidence | `unmeasured` |
+| Vision-language quality | Not applicable (vision disabled for this pack) |
+| Speech-recognition quality | Not applicable (audio disabled for this pack) |
+| Long-context quality | 262,144-token capacity is config metadata, not a validated claim |
+| Release certification | **Checkpoint Tier 1 certified** on `df-macstudio-m2` (2026-08-18), Hub commit `000000000000`; the formal AXQuant M0-M8 release campaign is a separate process and is not implied |
+
+## Modalities (capability-gated)
+
+Text checkpoint Tier 1 does **not** imply vision or audio quality. `Vision present=true` on a pack is not a quality pass.
+
+| Modality | Claim | Supported | Reason |
+| --- | --- | --- | --- |
+| Vision | `not-applicable` | `false` | vision not supported on this pack |
+| Audio | `not-applicable` | `false` | audio not supported on this pack |
+
+## Intended use and limitations
+
+- Intended for local development and evaluation on Apple Silicon with MLX-compatible runtimes.
+- No minimum unified-memory figure is claimed; loadability depends on model size, context length,
+  KV-cache policy, runtime buffers, and other processes using unified memory.
+- Architecture-prior allocation is not measured sensitivity. It must not be presented as measured
+  model quality.
+- The configured context window can require substantially more memory as the KV cache grows.
+- AX Engine execution is not established because this package has no validated native manifest.
+
+- Upstream capabilities, limitations, biases, and responsible-use guidance still apply.
+
+## Provenance and audit files
+
+- [`axquant_manifest.json`](axquant_manifest.json): package identity, byte accounting, runtime
+  contract, software versions, and file checksums.
+- [`axquant_plan.json`](axquant_plan.json): per-tensor precision decisions and planning evidence.
+- [`axquant_quantizer_execution.json`](axquant_quantizer_execution.json): conversion coverage and
+  fallback records.
+- [`axquant_runtime.json`](axquant_runtime.json): declared AX Engine and MLX compatibility metadata; runtime checks remain separate evidence.
+
+All published provenance uses repository-relative paths. Local source paths are stripped before
+publication. The checkpoint was converted from BF16 rather than re-quantized from an OptiQ
+artifact. If an OptiQ repository is published separately, it uses a different quantizer and
+should not be assumed to have identical BPW or quality.
+
+## License
+
+The checkpoint follows the upstream model license where applicable (often Apache License 2.0). See
+the [Qwen/Qwen3-Coder-Next model card](https://huggingface.co/Qwen/Qwen3-Coder-Next/tree/a7fbcb5c0e12d62a448eaa0e260346bf5dcc0feb) for license terms, model
+limitations, and responsible-use guidance.

axquant_manifest.json

@@ -0,0 +1,231 @@
+{
+  "axquant_version": "1.9.0",
+  "calibration": null,
+  "created_at": "2026-08-18T18:30:00.469669Z",
+  "effective_bpw": 5.059036550784871,
+  "files": [
+    {
+      "path": "README.md",
+      "sha256": "11e3eb7edd3b80c27700c00ab437368e0abdd4718d3109f8585abca977c6c74c",
+      "size_bytes": 8683
+    },
+    {
+      "path": "axquant_plan.json",
+      "sha256": "a83a02c5cec9f5065654fcdb071bf4d48ee74370ebbff2f8ac9377834866c65b",
+      "size_bytes": 67900706
+    },
+    {
+      "path": "axquant_quantizer_execution.json",
+      "sha256": "a7b7024c8a74a126aeead3162b55f4b923939b474c4b88a170138b5584281414",
+      "size_bytes": 18745977
+    },
+    {
+      "path": "axquant_runtime.json",
+      "sha256": "06d0b8ad801cc8d8a7d4ed799848088fec2381c110cf98d1da5ae548b6bfb30d",
+      "size_bytes": 1628
+    },
+    {
+      "path": "chat_template.jinja",
+      "sha256": "c79a833039a43602150cce0902403d6e376c50930c1b2a139b2964e1f0c322a0",
+      "size_bytes": 6068
+    },
+    {
+      "path": "config.json",
+      "sha256": "fc1e86652df25bc7dc455035a33dbd02c09deb91908ec2aad33f20e929dab6cf",
+      "size_bytes": 156950
+    },
+    {
+      "path": "generation_config.json",
+      "sha256": "37a3c1ef63516ca489c575f0db1c0405ddc0c3dbaca9ed987344c966c37aeef5",
+      "size_bytes": 214
+    },
+    {
+      "path": "model-00001-of-00009.safetensors",
+      "sha256": "25e0609df75f20e2d00380d92950b49ee112de291234a77a96bc222d4c1e5f4e",
+      "size_bytes": 5298640091
+    },
+    {
+      "path": "model-00002-of-00009.safetensors",
+      "sha256": "3d6a034a16a0a9fc1681d0654620b7aaf86781bd4b50b2730905d5f51585f1e3",
+      "size_bytes": 5251437577
+    },
+    {
+      "path": "model-00003-of-00009.safetensors",
+      "sha256": "1cac58ec604e6fa9255d6541f388056b3d505dca44c5d82c7b2f75d5267f1477",
+      "size_bytes": 5254914577
+    },
+    {
+      "path": "model-00004-of-00009.safetensors",
+      "sha256": "1f5d1903eed4f9b1785174a28f13cc870de4f765d2484ed4b84764b1371e9e51",
+      "size_bytes": 5251437678
+    },
+    {
+      "path": "model-00005-of-00009.safetensors",
+      "sha256": "25117fa9ceebe5dcd156850b540bc76da9d30c53bbbf2b367ae244ae302a861f",
+      "size_bytes": 5254914611
+    },
+    {
+      "path": "model-00006-of-00009.safetensors",
+      "sha256": "b4f5a246c38f81d5cad012bac46de7cb2e9900a16dad48209677482216dfc303",
+      "size_bytes": 5251437708
+    },
+    {
+      "path": "model-00007-of-00009.safetensors",
+      "sha256": "625f136ddd98f53968cf4e34c217e3a95e36c32d9b593d236c0fad683e73aae6",
+      "size_bytes": 5254914607
+    },
+    {
+      "path": "model-00008-of-00009.safetensors",
+      "sha256": "74234c80e06ab56f4d8af0918e8a3cd41223e6b1bfa210f369488b7b6e74e30b",
+      "size_bytes": 5251437720
+    },
+    {
+      "path": "model-00009-of-00009.safetensors",
+      "sha256": "dfb10b70f81105971143f8d0ee3ae65fd435604e4d0b428d3df3cd0e77a45a52",
+      "size_bytes": 909221492
+    },
+    {
+      "path": "model.safetensors.index.json",
+      "sha256": "7381ca01ebfa27aa1b45668f5bcc2c4bb2b6e53de49fd6763a04844c9cfd116a",
+      "size_bytes": 131598
+    },
+    {
+      "path": "qwen3_coder_detector_sgl.py",
+      "sha256": "e5a6112dd41df1580ede5f52e928e58cb74985dfa5628fd94073fd6a0f20fb8d",
+      "size_bytes": 19855
+    },
+    {
+      "path": "qwen3coder_tool_parser_vllm.py",
+      "sha256": "e6ad932aaa29a4004d592ca655c3d1ad849525a498ba60a716909ec9c1fda297",
+      "size_bytes": 31680
+    },
+    {
+      "path": "tokenizer.json",
+      "sha256": "be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506",
+      "size_bytes": 11422650
+    },
+    {
+      "path": "tokenizer_config.json",
+      "sha256": "23ff856b54c821dbd7e3fbcb9f33bd514f03a1ce194390d51bebae7de5a601be",
+      "size_bytes": 731
+    }
+  ],
+  "format": "mlx",
+  "logical_parameters": 79674391296,
+  "main_logical_parameters": 79674391296,
+  "main_weight_file_size_bytes": 42978356061,
+  "measured_main_bpw": 4.31539975260861,
+  "measured_total_bpw": 4.31539975260861,
+  "mtp_acceptance_retention": null,
+  "mtp_distribution": {},
+  "mtp_measured_speedup": null,
+  "mtp_policy": {
+    "candidate_bits": [
+      8,
+      16
+    ],
+    "min_bits": 8,
+    "mode": "protected",
+    "optimize_for_acceptance": true,
+    "preserve_external_sidecar": true,
+    "protect_norms": true,
+    "protect_output_head": true
+  },
+  "mtp_present": false,
+  "mtp_weight_file_size_bytes": 0,
+  "plan_sha256": "02567ac858019017a3421bcf536f3b748a1491abd64cd8416bda2942568023f1",
+  "profile": "agent-coding",
+  "protected_weight_file_size_bytes": 0,
+  "quantizer": "axquant",
+  "runtime": {
+    "ax_engine": {
+      "decode_kernel": null,
+      "fused_mtp": null,
+      "kernel_evidence": "unmeasured",
+      "model_manifest": "model-manifest.json",
+      "preferred_group_size": 32
+    },
+    "compatible_runtimes": [
+      {
+        "compatibility_level": "B",
+        "manifest": "config.json",
+        "mtp_support": "none",
+        "name": "mlx-lm",
+        "notes": [
+          "Standard backbone inference is the compatibility target.",
+          "AXQuant MTP metadata may be ignored by MLX-LM."
+        ],
+        "standard_inference": true,
+        "standard_mlx_weights": true,
+        "support_level": "standard-inference"
+      }
+    ],
+    "created_at": "2026-08-18T18:27:54.487021Z",
+    "kv_cache": null,
+    "memory_policy": {
+      "kv_cache_precision": "runtime-default",
+      "mtp_buffers": "not-required",
+      "prefix_cache": "runtime-managed",
+      "unified_memory_safety_margin": "benchmark-required"
+    },
+    "mtp": {
+      "acceptance_retention": null,
+      "detected": false,
+      "draft_tokens": null,
+      "enabled_by_default": false,
+      "head_precision": null,
+      "measured_speedup": null,
+      "optimized": false,
+      "recommended_temperature_max": null,
+      "sidecar_file": null,
+      "verification_mode": null
+    },
+    "optimization_scope": "text-path",
+    "primary_runtime": {
+      "compatibility_level": "A",
+      "manifest": "model-manifest.json",
+      "mtp_support": "none",
+      "name": "ax-engine",
+      "notes": [
+        "Runtime claims require a passing AX Engine doctor and benchmark report."
+      ],
+      "standard_inference": true,
+      "standard_mlx_weights": true,
+      "support_level": "optimized"
+    },
+    "schema_version": "axquant.runtime.v1"
+  },
+  "schema_version": "axquant.artifact.v2",
+  "software_versions": {
+    "ax_engine": null,
+    "axquant": "1.9.0",
+    "mlx": "0.32.0",
+    "mlx_lm": "0.31.3",
+    "pydantic": "2.13.4",
+    "python": "3.12.13",
+    "safetensors": "0.8.0"
+  },
+  "source_model": {
+    "architecture": "Qwen3NextForCausalLM",
+    "format": "mlx",
+    "local_path": null,
+    "model_id": "Qwen/Qwen3-Coder-Next",
+    "revision": "a7fbcb5c0e12d62a448eaa0e260346bf5dcc0feb"
+  },
+  "target_class": "5p6bpw",
+  "weight_distribution": {
+    "4bit": {
+      "fraction": 0.9915386718739344,
+      "parameters": 79000240128
+    },
+    "8bit": {
+      "fraction": 0.0045384078135800405,
+      "parameters": 361594880
+    },
+    "bf16": {
+      "fraction": 0.003922920312485546,
+      "parameters": 312556288
+    }
+  },
+  "weight_file_size_bytes": 42978356061
+}

axquant_plan.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a83a02c5cec9f5065654fcdb071bf4d48ee74370ebbff2f8ac9377834866c65b
+size 67900706

axquant_quantizer_execution.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7b7024c8a74a126aeead3162b55f4b923939b474c4b88a170138b5584281414
+size 18745977

axquant_runtime.json

@@ -0,0 +1,58 @@
+{
+  "ax_engine": {
+    "decode_kernel": null,
+    "fused_mtp": null,
+    "kernel_evidence": "unmeasured",
+    "model_manifest": "model-manifest.json",
+    "preferred_group_size": 32
+  },
+  "compatible_runtimes": [
+    {
+      "compatibility_level": "B",
+      "manifest": "config.json",
+      "mtp_support": "none",
+      "name": "mlx-lm",
+      "notes": [
+        "Standard backbone inference is the compatibility target.",
+        "AXQuant MTP metadata may be ignored by MLX-LM."
+      ],
+      "standard_inference": true,
+      "standard_mlx_weights": true,
+      "support_level": "standard-inference"
+    }
+  ],
+  "created_at": "2026-08-18T18:27:54.487021Z",
+  "kv_cache": null,
+  "memory_policy": {
+    "kv_cache_precision": "runtime-default",
+    "mtp_buffers": "not-required",
+    "prefix_cache": "runtime-managed",
+    "unified_memory_safety_margin": "benchmark-required"
+  },
+  "mtp": {
+    "acceptance_retention": null,
+    "detected": false,
+    "draft_tokens": null,
+    "enabled_by_default": false,
+    "head_precision": null,
+    "measured_speedup": null,
+    "optimized": false,
+    "recommended_temperature_max": null,
+    "sidecar_file": null,
+    "verification_mode": null
+  },
+  "optimization_scope": "text-path",
+  "primary_runtime": {
+    "compatibility_level": "A",
+    "manifest": "model-manifest.json",
+    "mtp_support": "none",
+    "name": "ax-engine",
+    "notes": [
+      "Runtime claims require a passing AX Engine doctor and benchmark report."
+    ],
+    "standard_inference": true,
+    "standard_mlx_weights": true,
+    "support_level": "optimized"
+  },
+  "schema_version": "axquant.runtime.v1"
+}

chat_template.jinja

@@ -0,0 +1,117 @@
+{% macro render_extra_keys(json_dict, handled_keys) %}
+    {%- if json_dict is mapping %}
+        {%- for json_key in json_dict if json_key not in handled_keys %}
+            {%- if json_dict[json_key] is string %}
+                {{-'\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | string) ~ '</' ~ json_key ~ '>' }}
+            {%- else %}
+                {{- '\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | tojson | safe) ~ '</' ~ json_key ~ '>' }}
+            {%- endif %}
+        {%- endfor %}
+    {%- endif %}
+{%- endmacro %}
+
+{%- if messages[0]["role"] == "system" %}
+    {%- set system_message = messages[0]["content"] %}
+    {%- set loop_messages = messages[1:] %}
+{%- else %}
+    {%- set loop_messages = messages %}
+{%- endif %}
+
+{%- if not tools is defined %}
+    {%- set tools = [] %}
+{%- endif %}
+
+{%- if system_message is defined %}
+    {{- "<|im_start|>system\n" + system_message }}
+{%- else %}
+    {%- if tools is iterable and tools | length > 0 %}
+        {{- "<|im_start|>system\nYou are Qwen, a helpful AI assistant that can interact with a computer to solve tasks." }}
+    {%- endif %}
+{%- endif %}
+{%- if tools is iterable and tools | length > 0 %}
+    {{- "\n\n# Tools\n\nYou have access to the following functions:\n\n" }}
+    {{- "<tools>" }}
+    {%- for tool in tools %}
+        {%- if tool.function is defined %}
+            {%- set tool = tool.function %}
+        {%- endif %}
+        {{- "\n<function>\n<name>" ~ tool.name ~ "</name>" }}
+        {%- if tool.description is defined %}
+            {{- '\n<description>' ~ (tool.description | trim) ~ '</description>' }}
+        {%- endif %}
+        {{- '\n<parameters>' }}
+        {%- if tool.parameters is defined and tool.parameters is mapping and tool.parameters.properties is defined and tool.parameters.properties is mapping %}
+            {%- for param_name, param_fields in tool.parameters.properties|items %}
+                {{- '\n<parameter>' }}
+                {{- '\n<name>' ~ param_name ~ '</name>' }}
+                {%- if param_fields.type is defined %}
+                    {{- '\n<type>' ~ (param_fields.type | string) ~ '</type>' }}
+                {%- endif %}
+                {%- if param_fields.description is defined %}
+                    {{- '\n<description>' ~ (param_fields.description | trim) ~ '</description>' }}
+                {%- endif %}
+                {%- set handled_keys = ['name', 'type', 'description'] %}
+                {{- render_extra_keys(param_fields, handled_keys) }}
+                {{- '\n</parameter>' }}
+            {%- endfor %}
+        {%- endif %}
+        {%- set handled_keys = ['type', 'properties'] %}
+        {{- render_extra_keys(tool.parameters, handled_keys) }}
+        {{- '\n</parameters>' }}
+        {%- set handled_keys = ['type', 'name', 'description', 'parameters'] %}
+        {{- render_extra_keys(tool, handled_keys) }}
+        {{- '\n</function>' }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+{%- endif %}
+{%- if system_message is defined %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if tools is iterable and tools | length > 0 %}
+        {{- '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in loop_messages %}
+    {%- if message.role == "assistant" and message.tool_calls is defined and message.tool_calls is iterable and message.tool_calls | length > 0 %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content is defined and message.content is string and message.content | trim | length > 0 %}
+            {{- '\n' + message.content | trim + '\n' }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+            {%- if tool_call.arguments is defined %}
+                {%- for args_name, args_value in tool_call.arguments|items %}
+                    {{- '<parameter=' + args_name + '>\n' }}
+                    {%- set args_value = args_value if args_value is string else args_value | tojson | safe %}
+                    {{- args_value }}
+                    {{- '\n</parameter>\n' }}
+                {%- endfor %}
+            {%- endif %}
+            {{- '</function>\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "user" or message.role == "system" or message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "temperature": 1.0,
+  "top_k": 40,
+  "top_p": 0.95,
+  "transformers_version": "4.57.3"
+}

model-00001-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25e0609df75f20e2d00380d92950b49ee112de291234a77a96bc222d4c1e5f4e
+size 5298640091

model-00002-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d6a034a16a0a9fc1681d0654620b7aaf86781bd4b50b2730905d5f51585f1e3
+size 5251437577

model-00003-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cac58ec604e6fa9255d6541f388056b3d505dca44c5d82c7b2f75d5267f1477
+size 5254914577

model-00004-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f5d1903eed4f9b1785174a28f13cc870de4f765d2484ed4b84764b1371e9e51
+size 5251437678

model-00005-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25117fa9ceebe5dcd156850b540bc76da9d30c53bbbf2b367ae244ae302a861f
+size 5254914611

model-00006-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4f5a246c38f81d5cad012bac46de7cb2e9900a16dad48209677482216dfc303
+size 5251437708

model-00007-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:625f136ddd98f53968cf4e34c217e3a95e36c32d9b593d236c0fad683e73aae6
+size 5254914607

model-00008-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74234c80e06ab56f4d8af0918e8a3cd41223e6b1bfa210f369488b7b6e74e30b
+size 5251437720

model-00009-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfb10b70f81105971143f8d0ee3ae65fd435604e4d0b428d3df3cd0e77a45a52
+size 909221492

qwen3_coder_detector_sgl.py

@@ -0,0 +1,474 @@
+import ast
+import json
+import logging
+import re
+from typing import Any, List, Optional
+
+from sglang.srt.entrypoints.openai.protocol import Tool
+from sglang.srt.function_call.base_format_detector import BaseFormatDetector
+from sglang.srt.function_call.core_types import (
+    StreamingParseResult,
+    ToolCallItem,
+    _GetInfoFunc,
+)
+
+logger = logging.getLogger(__name__)
+
+
+class Qwen3CoderDetector(BaseFormatDetector):
+    def __init__(self):
+        super().__init__()
+
+        # Sentinel tokens
+        self.tool_call_start_token: str = "<tool_call>"
+        self.tool_call_end_token: str = "</tool_call>"
+        self.tool_call_prefix: str = "<function="
+        self.function_end_token: str = "</function>"
+        self.parameter_prefix: str = "<parameter="
+        self.parameter_end_token: str = "</parameter>"
+
+        # Regex for non-streaming fallback
+        self.tool_call_regex = re.compile(r"<tool_call>(.*?)</tool_call>", re.DOTALL)
+        self.tool_call_function_regex = re.compile(
+            r"<function=(.*?)</function>|<function=(.*)$", re.DOTALL
+        )
+        self.tool_call_parameter_regex = re.compile(
+            r"<parameter=(.*?)(?:</parameter>|(?=<parameter=)|(?=</function>)|$)",
+            re.DOTALL,
+        )
+
+        # Streaming State
+        # Base class already initializes _buffer, we just use it directly
+        # No need to check with hasattr - we control the lifecycle through inheritance
+
+        # Index pointing to the next character to be processed in buffer
+        self.parsed_pos: int = 0
+        # Parameter count inside the current tool being processed, used to determine whether to add comma
+        self.current_tool_param_count: int = 0
+        # Flag indicating whether current tool has already sent '{'
+        self.json_started: bool = False
+
+        # [FIX] New state flag: mark whether inside tool_call structure block
+        self.is_inside_tool_call: bool = False
+
+        # Initialize attributes that were missing in the original PR
+        self.current_func_name: Optional[str] = None
+
+    def has_tool_call(self, text: str) -> bool:
+        return self.tool_call_start_token in text
+
+    def _get_arguments_config(
+        self, func_name: str, tools: Optional[list[Tool]]
+    ) -> dict:
+        """Extract argument configuration for a function."""
+        if tools is None:
+            return {}
+        for config in tools:
+            try:
+                config_type = config.type
+                config_function = config.function
+                config_function_name = config_function.name
+            except AttributeError:
+                continue
+
+            if config_type == "function" and config_function_name == func_name:
+                try:
+                    params = config_function.parameters
+                except AttributeError:
+                    return {}
+
+                if isinstance(params, dict) and "properties" in params:
+                    return params["properties"]
+                elif isinstance(params, dict):
+                    return params
+                else:
+                    return {}
+        logger.warning(f"Tool '{func_name}' is not defined in the tools list.")
+        return {}
+
+    def _convert_param_value(
+        self, param_value: str, param_name: str, param_config: dict, func_name: str
+    ) -> Any:
+        """Convert parameter value based on its type in the schema."""
+        # Handle null value for any type
+        if param_value.lower() == "null":
+            return None
+
+        if param_name not in param_config:
+            if param_config != {}:
+                logger.warning(
+                    f"Parsed parameter '{param_name}' is not defined in the tool "
+                    f"parameters for tool '{func_name}', directly returning the string value."
+                )
+            return param_value
+
+        if (
+            isinstance(param_config[param_name], dict)
+            and "type" in param_config[param_name]
+        ):
+            param_type = str(param_config[param_name]["type"]).strip().lower()
+        else:
+            param_type = "string"
+        if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
+            return param_value
+        elif (
+            param_type.startswith("int")
+            or param_type.startswith("uint")
+            or param_type.startswith("long")
+            or param_type.startswith("short")
+            or param_type.startswith("unsigned")
+        ):
+            try:
+                param_value = int(param_value)
+            except Exception:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' is not an integer in tool "
+                    f"'{func_name}', degenerating to string."
+                )
+            return param_value
+        elif param_type.startswith("num") or param_type.startswith("float"):
+            try:
+                maybe_convert = (
+                    False if "." in param_value or "e" in param_value.lower() else True
+                )
+                param_value: float = float(param_value)
+                if maybe_convert and param_value.is_integer():
+                    param_value = int(param_value)
+            except Exception:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' is not a float in tool "
+                    f"'{func_name}', degenerating to string."
+                )
+            return param_value
+        elif param_type in ["boolean", "bool", "binary"]:
+            param_value = param_value.lower()
+            if param_value not in ["true", "false"]:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' is not a boolean (`true` of `false`) in tool '{func_name}', degenerating to false."
+                )
+            return param_value == "true"
+        else:
+            if (
+                param_type in ["object", "array", "arr"]
+                or param_type.startswith("dict")
+                or param_type.startswith("list")
+            ):
+                try:
+                    param_value = json.loads(param_value)
+                    return param_value
+                except Exception:
+                    logger.warning(
+                        f"Parsed value '{param_value}' of parameter '{param_name}' cannot be parsed with json.loads in tool "
+                        f"'{func_name}', will try other methods to parse it."
+                    )
+            try:
+                param_value = ast.literal_eval(param_value)  # safer
+            except Exception:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' cannot be converted via Python `ast.literal_eval()` in tool '{func_name}', degenerating to string."
+                )
+            return param_value
+
+    def detect_and_parse(self, text: str, tools: List[Tool]) -> StreamingParseResult:
+        """One-shot parsing for non-streaming scenarios."""
+        if self.tool_call_start_token not in text:
+            return StreamingParseResult(normal_text=text)
+
+        calls = []
+        try:
+            # Simple cleanup of the text to find tool calls
+            # Note: This is a simplified regex approach consistent with vLLM
+            raw_tool_calls = self.tool_call_regex.findall(text)
+            if not raw_tool_calls:
+                # Fallback: maybe the whole text is inside the tag or tags are stripped
+                if self.tool_call_prefix in text:
+                    raw_tool_calls = [text]
+
+            tool_idx = 0
+            for tool_content in raw_tool_calls:
+                # Find function calls
+                funcs = self.tool_call_function_regex.findall(tool_content)
+                for func_match in funcs:
+                    func_body = func_match[0] or func_match[1]
+                    if ">" not in func_body:
+                        continue
+
+                    name_end = func_body.index(">")
+                    func_name = func_body[:name_end]
+                    params_str = func_body[name_end + 1 :]
+
+                    param_config = self._get_arguments_config(func_name, tools)
+                    parsed_params = {}
+
+                    for p_match in self.tool_call_parameter_regex.findall(params_str):
+                        if ">" not in p_match:
+                            continue
+                        p_idx = p_match.index(">")
+                        p_name = p_match[:p_idx]
+                        p_val = p_match[p_idx + 1 :]
+                        # Remove prefixing and trailing \n
+                        if p_val.startswith("\n"):
+                            p_val = p_val[1:]
+                        if p_val.endswith("\n"):
+                            p_val = p_val[:-1]
+
+                        parsed_params[p_name] = self._convert_param_value(
+                            p_val, p_name, param_config, func_name
+                        )
+
+                    calls.append(
+                        ToolCallItem(
+                            tool_index=tool_idx,
+                            name=func_name,
+                            parameters=json.dumps(parsed_params, ensure_ascii=False),
+                        )
+                    )
+                    tool_idx += 1
+
+            # Determine normal text (text before the first tool call)
+            start_idx = text.find(self.tool_call_start_token)
+            if start_idx == -1:
+                start_idx = text.find(self.tool_call_prefix)
+            normal_text = text[:start_idx] if start_idx > 0 else ""
+
+            return StreamingParseResult(normal_text=normal_text, calls=calls)
+
+        except Exception as e:
+            logger.error(f"Error in detect_and_parse: {e}")
+            return StreamingParseResult(normal_text=text)
+
+    def parse_streaming_increment(
+        self, new_text: str, tools: List[Tool]
+    ) -> StreamingParseResult:
+        """
+        Robust cursor-based streaming parser.
+        """
+        self._buffer += new_text
+
+        # Guard against empty buffer
+        if not self._buffer:
+            return StreamingParseResult()
+
+        calls = []
+        normal_text_chunks = []
+
+        while True:
+            # Working text slice
+            current_slice = self._buffer[self.parsed_pos :]
+
+            # Optimization: If almost empty, wait for more
+            if not current_slice:
+                break
+
+            # -------------------------------------------------------
+            # 1. Priority detection: check if it's the start of Tool Call
+            # -------------------------------------------------------
+            if current_slice.startswith(self.tool_call_start_token):
+                self.parsed_pos += len(self.tool_call_start_token)
+                self.is_inside_tool_call = True
+                continue
+
+            # -------------------------------------------------------
+            # 2. Function Name: <function=name>
+            # -------------------------------------------------------
+            if current_slice.startswith(self.tool_call_prefix):
+                end_angle = current_slice.find(">")
+                if end_angle != -1:
+                    func_name = current_slice[len(self.tool_call_prefix) : end_angle]
+
+                    self.current_tool_id += 1
+                    self.current_tool_name_sent = True
+                    self.current_tool_param_count = 0
+                    self.json_started = False
+                    self.current_func_name = func_name
+
+                    calls.append(
+                        ToolCallItem(
+                            tool_index=self.current_tool_id,
+                            name=func_name,
+                            parameters="",
+                        )
+                    )
+
+                    self.parsed_pos += end_angle + 1
+                    continue
+                else:
+                    # Incomplete tag
+                    break
+
+            # -------------------------------------------------------
+            # 3. Parameter: <parameter=name>value...
+            # -------------------------------------------------------
+            if current_slice.startswith(self.parameter_prefix):
+                name_end = current_slice.find(">")
+                if name_end != -1:
+                    value_start_idx = name_end + 1
+                    rest_of_slice = current_slice[value_start_idx:]
+
+                    # A parameter can end in multiple ways:
+                    # 1. [Normal] Encounter </parameter>
+                    # 2. [Abnormal] Encounter next <parameter=
+                    # 3. [Abnormal] Encounter </function>
+                    # So we need to find the smallest one as the parameter end position.
+                    cand_end_param = rest_of_slice.find(self.parameter_end_token)
+                    cand_next_param = rest_of_slice.find(self.parameter_prefix)
+                    cand_end_func = rest_of_slice.find(self.function_end_token)
+
+                    candidates = []
+                    if cand_end_param != -1:
+                        candidates.append(
+                            (cand_end_param, len(self.parameter_end_token))
+                        )
+                    if cand_next_param != -1:
+                        candidates.append((cand_next_param, 0))
+                    if cand_end_func != -1:
+                        candidates.append((cand_end_func, 0))
+
+                    if candidates:
+                        best_cand = min(candidates, key=lambda x: x[0])
+                        end_pos = best_cand[0]
+                        end_token_len = best_cand[1]
+
+                        param_name = current_slice[
+                            len(self.parameter_prefix) : name_end
+                        ]
+                        raw_value = rest_of_slice[:end_pos]
+
+                        # Cleanup value
+                        if raw_value.startswith("\n"):
+                            raw_value = raw_value[1:]
+                        if raw_value.endswith("\n"):
+                            raw_value = raw_value[:-1]
+
+                        # JSON Construction
+                        if not self.json_started:
+                            calls.append(
+                                ToolCallItem(
+                                    tool_index=self.current_tool_id, parameters="{"
+                                )
+                            )
+                            self.json_started = True
+
+                        param_config = self._get_arguments_config(
+                            self.current_func_name, tools
+                        )
+                        converted_val = self._convert_param_value(
+                            raw_value, param_name, param_config, self.current_func_name
+                        )
+
+                        # Construct JSON fragment: "key": value
+                        # Note: We must be careful with json.dumps to ensure valid JSON streaming
+                        json_key_val = f"{json.dumps(param_name)}: {json.dumps(converted_val, ensure_ascii=False)}"
+
+                        if self.current_tool_param_count > 0:
+                            fragment = f", {json_key_val}"
+                        else:
+                            fragment = json_key_val
+
+                        calls.append(
+                            ToolCallItem(
+                                tool_index=self.current_tool_id, parameters=fragment
+                            )
+                        )
+                        self.current_tool_param_count += 1
+
+                        # Advance cursor
+                        total_len = (name_end + 1) + end_pos + end_token_len
+                        self.parsed_pos += total_len
+                        continue
+
+                # Incomplete parameter tag or value
+                break
+
+            # -------------------------------------------------------
+            # 4. Function End: </function>
+            # -------------------------------------------------------
+            if current_slice.startswith(self.function_end_token):
+                if not self.json_started:
+                    calls.append(
+                        ToolCallItem(tool_index=self.current_tool_id, parameters="{")
+                    )
+                    self.json_started = True
+
+                calls.append(
+                    ToolCallItem(tool_index=self.current_tool_id, parameters="}")
+                )
+                self.parsed_pos += len(self.function_end_token)
+                self.current_func_name = None
+                continue
+
+            # -------------------------------------------------------
+            # 5. Tool Call End: </tool_call>
+            # -------------------------------------------------------
+            if current_slice.startswith(self.tool_call_end_token):
+                self.parsed_pos += len(self.tool_call_end_token)
+                self.is_inside_tool_call = False  # [FIX] Exit tool call region
+                continue
+
+            # -------------------------------------------------------
+            # 6. Handling content / whitespace / normal text
+            # -------------------------------------------------------
+            # If current position is not the start of a tag (i.e., doesn't start with <), it might be plain text,
+            # or a newline between two tags.
+            # But we need to be careful not to output truncated tags like "<fun" as text.
+
+            next_open_angle = current_slice.find("<")
+
+            if next_open_angle == -1:
+                # This entire segment is plain text
+                if not self.is_inside_tool_call:
+                    normal_text_chunks.append(current_slice)
+                # [FIX] If inside tool call, discard this text (usually \n), don't append
+                self.parsed_pos += len(current_slice)
+                continue
+
+            elif next_open_angle == 0:
+                # Looks like a Tag, but doesn't match any known Tag above
+
+                possible_tags = [
+                    self.tool_call_start_token,
+                    self.tool_call_end_token,
+                    self.tool_call_prefix,
+                    self.function_end_token,
+                    self.parameter_prefix,
+                    self.parameter_end_token,
+                ]
+
+                is_potential_tag = False
+                for tag in possible_tags:
+                    if tag.startswith(current_slice):
+                        is_potential_tag = True
+                        break
+
+                if is_potential_tag:
+                    break  # Wait for more
+                else:
+                    # Just a plain '<' symbol
+                    if not self.is_inside_tool_call:
+                        normal_text_chunks.append("<")
+                    self.parsed_pos += 1
+                    continue
+
+            else:
+                # '<' is in the middle
+                text_segment = current_slice[:next_open_angle]
+                if not self.is_inside_tool_call:
+                    normal_text_chunks.append(text_segment)
+                # [FIX] If inside tool call, discard whitespace/text before Tag
+                self.parsed_pos += next_open_angle
+                continue
+
+        # Memory Cleanup: Slice the buffer
+        # Keep unparsed part, discard parsed part
+        if self.parsed_pos > 0:
+            self._buffer = self._buffer[self.parsed_pos :]
+            self.parsed_pos = 0
+
+        normal_text = "".join(normal_text_chunks) if normal_text_chunks else ""
+        return StreamingParseResult(calls=calls, normal_text=normal_text)
+
+    def supports_structural_tag(self) -> bool:
+        return False
+
+    def structure_info(self) -> _GetInfoFunc:
+        raise NotImplementedError

qwen3coder_tool_parser_vllm.py

@@ -0,0 +1,690 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+import ast
+import json
+import uuid
+from collections.abc import Sequence
+from typing import Any, List, Optional, Union
+
+import regex as re
+
+from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
+                                              ChatCompletionToolsParam,
+                                              DeltaFunctionCall, DeltaMessage,
+                                              DeltaToolCall,
+                                              ExtractedToolCallInformation,
+                                              FunctionCall, ToolCall)
+from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
+    ToolParser, ToolParserManager)
+from vllm.logger import init_logger
+from vllm.transformers_utils.tokenizer import AnyTokenizer
+
+logger = init_logger(__name__)
+
+
+@ToolParserManager.register_module("qwen3_coder")
+class Qwen3CoderToolParser(ToolParser):
+
+    def __init__(self, tokenizer: AnyTokenizer):
+        super().__init__(tokenizer)
+
+        self.current_tool_name_sent: bool = False
+        self.prev_tool_call_arr: list[dict] = []
+        self.current_tool_id: int = -1
+        self.streamed_args_for_tool: list[str] = []
+
+        # Sentinel tokens for streaming mode
+        self.tool_call_start_token: str = "<tool_call>"
+        self.tool_call_end_token: str = "</tool_call>"
+        self.tool_call_prefix: str = "<function="
+        self.function_end_token: str = "</function>"
+        self.parameter_prefix: str = "<parameter="
+        self.parameter_end_token: str = "</parameter>"
+        self.is_tool_call_started: bool = False
+        self.failed_count: int = 0
+
+        # Enhanced streaming state - reset for each new message
+        self._reset_streaming_state()
+
+        # Regex patterns
+        self.tool_call_complete_regex = re.compile(
+            r"<tool_call>(.*?)</tool_call>", re.DOTALL)
+        self.tool_call_regex = re.compile(
+            r"<tool_call>(.*?)</tool_call>|<tool_call>(.*?)$", re.DOTALL)
+        self.tool_call_function_regex = re.compile(
+            r"<function=(.*?)</function>|<function=(.*)$", re.DOTALL)
+        self.tool_call_parameter_regex = re.compile(
+            r"<parameter=(.*?)(?:</parameter>|(?=<parameter=)|(?=</function>)|$)",
+            re.DOTALL)
+
+        if not self.model_tokenizer:
+            raise ValueError(
+                "The model tokenizer must be passed to the ToolParser "
+                "constructor during construction.")
+
+        self.tool_call_start_token_id = self.vocab.get(
+            self.tool_call_start_token)
+        self.tool_call_end_token_id = self.vocab.get(self.tool_call_end_token)
+
+        if self.tool_call_start_token_id is None or self.tool_call_end_token_id is None:
+            raise RuntimeError(
+                "Qwen3 XML Tool parser could not locate tool call start/end "
+                "tokens in the tokenizer!")
+
+        logger.info(
+            f"vLLM Successfully import tool parser {self.__class__.__name__} !"
+        )
+
+    def _generate_tool_call_id(self) -> str:
+        """Generate a unique tool call ID."""
+        return f"call_{uuid.uuid4().hex[:24]}"
+
+    def _reset_streaming_state(self):
+        """Reset all streaming state."""
+        self.current_tool_index = 0
+        self.is_tool_call_started = False
+        self.header_sent = False
+        self.current_tool_id = None
+        self.current_function_name = None
+        self.current_param_name = None
+        self.current_param_value = ""
+        self.param_count = 0
+        self.in_param = False
+        self.in_function = False
+        self.accumulated_text = ""
+        self.json_started = False
+        self.json_closed = False
+        # Store accumulated parameters for type conversion
+        self.accumulated_params = {}
+        self.streaming_request = None
+
+    def _get_arguments_config(
+            self, func_name: str,
+            tools: Optional[list[ChatCompletionToolsParam]]) -> dict:
+        """Extract argument configuration for a function."""
+        if tools is None:
+            return {}
+        for config in tools:
+            if not hasattr(config, "type") or not (hasattr(
+                    config, "function") and hasattr(config.function, "name")):
+                continue
+            if config.type == "function" and config.function.name == func_name:
+                if not hasattr(config.function, "parameters"):
+                    return {}
+                params = config.function.parameters
+                if isinstance(params, dict) and "properties" in params:
+                    return params["properties"]
+                elif isinstance(params, dict):
+                    return params
+                else:
+                    return {}
+        logger.warning(f"Tool '{func_name}' is not defined in the tools list.")
+        return {}
+
+    def _convert_param_value(self, param_value: str, param_name: str,
+                             param_config: dict, func_name: str) -> Any:
+        """Convert parameter value based on its type in the schema."""
+        # Handle null value for any type
+        if param_value.lower() == "null":
+            return None
+
+        if param_name not in param_config:
+            if param_config != {}:
+                logger.warning(
+                    f"Parsed parameter '{param_name}' is not defined in the tool "
+                    f"parameters for tool '{func_name}', directly returning the string value."
+                )
+            return param_value
+
+        if isinstance(param_config[param_name],
+                      dict) and "type" in param_config[param_name]:
+            param_type = str(param_config[param_name]["type"]).strip().lower()
+        else:
+            param_type = "string"
+        if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
+            return param_value
+        elif param_type.startswith("int") or param_type.startswith(
+                "uint") or param_type.startswith(
+                    "long") or param_type.startswith(
+                        "short") or param_type.startswith("unsigned"):
+            try:
+                param_value = int(param_value)
+            except:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' is not an integer in tool "
+                    f"'{func_name}', degenerating to string.")
+            return param_value
+        elif param_type.startswith("num") or param_type.startswith("float"):
+            try:
+                maybe_convert = False if "." in param_value or "e" in param_value.lower() else True
+                param_value: float = float(param_value)
+                if maybe_convert and param_value.is_integer():
+                    param_value = int(param_value)
+            except:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' is not a float in tool "
+                    f"'{func_name}', degenerating to string.")
+            return param_value
+        elif param_type in ["boolean", "bool", "binary"]:
+            param_value = param_value.lower()
+            if param_value not in ["true", "false"]:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' is not a boolean (`true` of `false`) in tool '{func_name}', degenerating to false."
+                )
+            return param_value == "true"
+        else:
+            if param_type in ["object", "array", "arr"
+                              ] or param_type.startswith(
+                                  "dict") or param_type.startswith("list"):
+                try:
+                    param_value = json.loads(param_value)
+                    return param_value
+                except:
+                    logger.warning(
+                        f"Parsed value '{param_value}' of parameter '{param_name}' cannot be parsed with json.loads in tool "
+                        f"'{func_name}', will try other methods to parse it.")
+            try:
+                param_value = ast.literal_eval(param_value)  # safer
+            except:
+                logger.warning(
+                    f"Parsed value '{param_value}' of parameter '{param_name}' cannot be converted via Python `ast.literal_eval()` in tool '{func_name}', degenerating to string."
+                )
+            return param_value
+
+    def _parse_xml_function_call(
+            self, function_call_str: str,
+            tools: Optional[list[ChatCompletionToolsParam]]
+    ) -> Optional[ToolCall]:
+
+        # Extract function name
+        end_index = function_call_str.index(">")
+        function_name = function_call_str[:end_index]
+        param_config = self._get_arguments_config(function_name, tools)
+        parameters = function_call_str[end_index + 1:]
+        param_dict = {}
+        for match_text in self.tool_call_parameter_regex.findall(parameters):
+            idx = match_text.index(">")
+            param_name = match_text[:idx]
+            param_value = str(match_text[idx + 1:])
+            # Remove prefix and trailing \n
+            if param_value.startswith("\n"):
+                param_value = param_value[1:]
+            if param_value.endswith("\n"):
+                param_value = param_value[:-1]
+
+            param_dict[param_name] = self._convert_param_value(
+                param_value, param_name, param_config, function_name)
+        return ToolCall(
+            type="function",
+            function=FunctionCall(name=function_name,
+                                  arguments=json.dumps(param_dict,
+                                                       ensure_ascii=False)),
+        )
+
+    def _get_function_calls(self, model_output: str) -> List[str]:
+        # Find all tool calls
+        matched_ranges = self.tool_call_regex.findall(model_output)
+        raw_tool_calls = [
+            match[0] if match[0] else match[1] for match in matched_ranges
+        ]
+
+        # Back-off strategy if no tool_call tags found
+        if len(raw_tool_calls) == 0:
+            raw_tool_calls = [model_output]
+
+        raw_function_calls = []
+        for tool_call in raw_tool_calls:
+            raw_function_calls.extend(
+                self.tool_call_function_regex.findall(tool_call))
+
+        function_calls = [
+            match[0] if match[0] else match[1] for match in raw_function_calls
+        ]
+        return function_calls
+
+    def extract_tool_calls(
+        self,
+        model_output: str,
+        request: ChatCompletionRequest,
+    ) -> ExtractedToolCallInformation:
+        # Quick check to avoid unnecessary processing
+        if self.tool_call_prefix not in model_output:
+            return ExtractedToolCallInformation(tools_called=False,
+                                                tool_calls=[],
+                                                content=model_output)
+
+        try:
+            function_calls = self._get_function_calls(model_output)
+            if len(function_calls) == 0:
+                return ExtractedToolCallInformation(tools_called=False,
+                                                    tool_calls=[],
+                                                    content=model_output)
+
+            tool_calls = [
+                self._parse_xml_function_call(function_call_str, request.tools)
+                for function_call_str in function_calls
+            ]
+
+            # Populate prev_tool_call_arr for serving layer to set finish_reason
+            self.prev_tool_call_arr.clear()  # Clear previous calls
+            for tool_call in tool_calls:
+                if tool_call:
+                    self.prev_tool_call_arr.append({
+                        "name":
+                        tool_call.function.name,
+                        "arguments":
+                        tool_call.function.arguments,
+                    })
+
+            # Extract content before tool calls
+            content_index = model_output.find(self.tool_call_start_token)
+            content_index = content_index if content_index >= 0 else model_output.find(
+                self.tool_call_prefix)
+            content = model_output[:content_index]  # .rstrip()
+
+            return ExtractedToolCallInformation(
+                tools_called=(len(tool_calls) > 0),
+                tool_calls=tool_calls,
+                content=content if content else None,
+            )
+
+        except Exception:
+            logger.exception("Error in extracting tool call from response.")
+            return ExtractedToolCallInformation(tools_called=False,
+                                                tool_calls=[],
+                                                content=model_output)
+
+    def extract_tool_calls_streaming(
+        self,
+        previous_text: str,
+        current_text: str,
+        delta_text: str,
+        previous_token_ids: Sequence[int],
+        current_token_ids: Sequence[int],
+        delta_token_ids: Sequence[int],
+        request: ChatCompletionRequest,
+    ) -> Union[DeltaMessage, None]:
+        # Store request for type conversion
+        if not previous_text:
+            self._reset_streaming_state()
+            self.streaming_request = request
+
+        # If no delta text, return None unless it's an EOS token after tool calls
+        if not delta_text:
+            # Check if this is an EOS token after all tool calls are complete
+            # We check for tool calls in the text even if is_tool_call_started is False
+            # because it might have been reset after processing all tools
+            if delta_token_ids and self.tool_call_end_token_id not in delta_token_ids:
+                # Count complete tool calls
+                complete_calls = len(
+                    self.tool_call_complete_regex.findall(current_text))
+
+                # If we have completed tool calls and populated prev_tool_call_arr
+                if complete_calls > 0 and len(self.prev_tool_call_arr) > 0:
+                    # Check if all tool calls are closed
+                    open_calls = current_text.count(
+                        self.tool_call_start_token) - current_text.count(
+                            self.tool_call_end_token)
+                    if open_calls == 0:
+                        # Return empty delta message to allow finish_reason processing
+                        return DeltaMessage(content="")
+                elif not self.is_tool_call_started and current_text:
+                    # This is a regular content response that's now complete
+                    return DeltaMessage(content="")
+            return None
+
+        # Update accumulated text
+        self.accumulated_text = current_text
+
+        # Check if we need to advance to next tool
+        if self.json_closed and not self.in_function:
+            # Check if this tool call has ended
+            tool_ends = current_text.count(self.tool_call_end_token)
+            if tool_ends > self.current_tool_index:
+                # This tool has ended, advance to next
+                self.current_tool_index += 1
+                self.header_sent = False
+                self.param_count = 0
+                self.json_started = False
+                self.json_closed = False
+                self.accumulated_params = {}
+
+                # Check if there are more tool calls
+                tool_starts = current_text.count(self.tool_call_start_token)
+                if self.current_tool_index >= tool_starts:
+                    # No more tool calls
+                    self.is_tool_call_started = False
+                # Continue processing next tool
+                return None
+
+        # Handle normal content before tool calls
+        if not self.is_tool_call_started:
+            # Check if tool call is starting
+            if self.tool_call_start_token_id in delta_token_ids or self.tool_call_start_token in delta_text:
+                self.is_tool_call_started = True
+                # Return any content before the tool call
+                if self.tool_call_start_token in delta_text:
+                    content_before = delta_text[:delta_text.index(
+                        self.tool_call_start_token)]
+                    if content_before:
+                        return DeltaMessage(content=content_before)
+                return None
+            else:
+                # Check if we're between tool calls - skip whitespace
+                if current_text.rstrip().endswith(self.tool_call_end_token):
+                    # We just ended a tool call, skip whitespace
+                    if delta_text.strip() == "":
+                        return None
+                # Normal content, no tool call
+                return DeltaMessage(content=delta_text)
+
+        # Check if we're between tool calls (waiting for next one)
+        # Count tool calls we've seen vs processed
+        tool_starts_count = current_text.count(self.tool_call_start_token)
+        if self.current_tool_index >= tool_starts_count:
+            # We're past all tool calls, shouldn't be here
+            return None
+
+        # We're in a tool call, find the current tool call portion
+        # Need to find the correct tool call based on current_tool_index
+        tool_starts = []
+        idx = 0
+        while True:
+            idx = current_text.find(self.tool_call_start_token, idx)
+            if idx == -1:
+                break
+            tool_starts.append(idx)
+            idx += len(self.tool_call_start_token)
+
+        if self.current_tool_index >= len(tool_starts):
+            # No more tool calls to process yet
+            return None
+
+        tool_start_idx = tool_starts[self.current_tool_index]
+        # Find where this tool call ends (or current position if not ended yet)
+        tool_end_idx = current_text.find(self.tool_call_end_token,
+                                         tool_start_idx)
+        if tool_end_idx == -1:
+            tool_text = current_text[tool_start_idx:]
+        else:
+            tool_text = current_text[tool_start_idx:tool_end_idx +
+                                     len(self.tool_call_end_token)]
+
+        # Looking for function header
+        if not self.header_sent:
+            if self.tool_call_prefix in tool_text:
+                func_start = tool_text.find(self.tool_call_prefix) + len(
+                    self.tool_call_prefix)
+                func_end = tool_text.find(">", func_start)
+
+                if func_end != -1:
+                    # Found complete function name
+                    self.current_function_name = tool_text[func_start:func_end]
+                    self.current_tool_id = self._generate_tool_call_id()
+                    self.header_sent = True
+                    self.in_function = True
+
+                    # IMPORTANT: Add to prev_tool_call_arr immediately when we detect a tool call
+                    # This ensures finish_reason="tool_calls" even if parsing isn't complete
+                    already_added = any(
+                        tool.get("name") == self.current_function_name
+                        for tool in self.prev_tool_call_arr)
+                    if not already_added:
+                        self.prev_tool_call_arr.append({
+                            "name": self.current_function_name,
+                            "arguments":
+                            "{}",  # Placeholder, will be updated later
+                        })
+
+                    # Send header with function info
+                    return DeltaMessage(tool_calls=[
+                        DeltaToolCall(
+                            index=self.current_tool_index,
+                            id=self.current_tool_id,
+                            function=DeltaFunctionCall(
+                                name=self.current_function_name, arguments=""),
+                            type="function",
+                        )
+                    ])
+            return None
+
+        # We've sent header, now handle function body
+        if self.in_function:
+            # Send opening brace if not sent yet
+            if not self.json_started and self.parameter_prefix not in delta_text:
+                self.json_started = True
+                return DeltaMessage(tool_calls=[
+                    DeltaToolCall(
+                        index=self.current_tool_index,
+                        function=DeltaFunctionCall(arguments="{"),
+                    )
+                ])
+
+            # Make sure json_started is set if we're processing parameters
+            if not self.json_started:
+                self.json_started = True
+
+            # Check for function end in accumulated text
+            if not self.json_closed and self.function_end_token in tool_text:
+                # Close JSON
+                self.json_closed = True
+
+                # Extract the complete tool call to update prev_tool_call_arr with final arguments
+                # Find the function content
+                func_start = tool_text.find(self.tool_call_prefix) + len(
+                    self.tool_call_prefix)
+                func_content_end = tool_text.find(self.function_end_token,
+                                                  func_start)
+                if func_content_end != -1:
+                    func_content = tool_text[func_start:func_content_end]
+                    # Parse to get the complete arguments
+                    try:
+                        parsed_tool = self._parse_xml_function_call(
+                            func_content, self.streaming_request.tools
+                            if self.streaming_request else None)
+                        if parsed_tool:
+                            # Update existing entry in prev_tool_call_arr with complete arguments
+                            for i, tool in enumerate(self.prev_tool_call_arr):
+                                if tool.get(
+                                        "name") == parsed_tool.function.name:
+                                    self.prev_tool_call_arr[i][
+                                        "arguments"] = parsed_tool.function.arguments
+                                    break
+                    except Exception:
+                        pass  # Ignore parsing errors during streaming
+
+                result = DeltaMessage(tool_calls=[
+                    DeltaToolCall(
+                        index=self.current_tool_index,
+                        function=DeltaFunctionCall(arguments="}"),
+                    )
+                ])
+
+                # Reset state for next tool
+                self.in_function = False
+                self.json_closed = True
+                self.accumulated_params = {}
+
+                return result
+
+            # Look for parameters
+            # Find all parameter starts
+            param_starts = []
+            idx = 0
+            while True:
+                idx = tool_text.find(self.parameter_prefix, idx)
+                if idx == -1:
+                    break
+                param_starts.append(idx)
+                idx += len(self.parameter_prefix)
+
+            # Check if we should start a new parameter
+            if not self.in_param and self.param_count < len(param_starts):
+
+                if len(param_starts) > self.param_count:
+                    # Process the next parameter
+                    param_idx = param_starts[self.param_count]
+                    param_start = param_idx + len(self.parameter_prefix)
+                    remaining = tool_text[param_start:]
+
+                    if ">" in remaining:
+                        # We have the complete parameter name
+                        name_end = remaining.find(">")
+                        self.current_param_name = remaining[:name_end]
+
+                        # Find the parameter value
+                        value_start = param_start + name_end + 1
+                        value_text = tool_text[value_start:]
+                        if value_text.startswith("\n"):
+                            value_text = value_text[1:]
+
+                        # Find where this parameter ends
+                        param_end_idx = value_text.find(
+                            self.parameter_end_token)
+                        if param_end_idx == -1:
+                            # No closing tag, look for next parameter or function end
+                            next_param_idx = value_text.find(
+                                self.parameter_prefix)
+                            func_end_idx = value_text.find(
+                                self.function_end_token)
+
+                            if next_param_idx != -1 and (func_end_idx == -1
+                                                         or next_param_idx
+                                                         < func_end_idx):
+                                param_end_idx = next_param_idx
+                            elif func_end_idx != -1:
+                                param_end_idx = func_end_idx
+                            else:
+                                # Neither found, check if tool call is complete
+                                if self.tool_call_end_token in tool_text:
+                                    # Tool call is complete, so parameter must be complete too
+                                    # Use all remaining text before function end as value
+                                    param_end_idx = len(value_text)
+                                else:
+                                    # Still streaming, wait for more content
+                                    return None
+
+                        if param_end_idx != -1:
+                            # Complete parameter found
+                            param_value = value_text[:param_end_idx]
+                            if param_value.endswith("\n"):
+                                param_value = param_value[:-1]
+
+                            # Store raw value for later processing
+                            self.accumulated_params[
+                                self.current_param_name] = param_value
+
+                            # Get parameter configuration for type conversion
+                            param_config = self._get_arguments_config(
+                                self.current_function_name,
+                                self.streaming_request.tools
+                                if self.streaming_request else None)
+
+                            # Convert the parameter value to the appropriate type
+                            converted_value = self._convert_param_value(
+                                param_value, self.current_param_name,
+                                param_config, self.current_function_name)
+
+                            # Build JSON fragment based on the converted type
+                            # Use json.dumps to properly serialize the value
+                            serialized_value = json.dumps(converted_value,
+                                                          ensure_ascii=False)
+
+                            if self.param_count == 0:
+                                json_fragment = f'"{self.current_param_name}": {serialized_value}'
+                            else:
+                                json_fragment = f', "{self.current_param_name}": {serialized_value}'
+
+                            self.param_count += 1
+
+                            return DeltaMessage(tool_calls=[
+                                DeltaToolCall(
+                                    index=self.current_tool_index,
+                                    function=DeltaFunctionCall(
+                                        arguments=json_fragment),
+                                )
+                            ])
+
+            # Continue parameter value - Not used in the current implementation
+            # since we process complete parameters above
+            if self.in_param:
+                if self.parameter_end_token in delta_text:
+                    # End of parameter
+                    end_idx = delta_text.find(self.parameter_end_token)
+                    value_chunk = delta_text[:end_idx]
+
+                    # Skip past > if at start
+                    if not self.current_param_value and ">" in value_chunk:
+                        gt_idx = value_chunk.find(">")
+                        value_chunk = value_chunk[gt_idx + 1:]
+
+                    if not self.current_param_value and value_chunk.startswith(
+                            "\n"):
+                        value_chunk = value_chunk[1:]
+
+                    # Store complete value
+                    full_value = self.current_param_value + value_chunk
+                    self.accumulated_params[
+                        self.current_param_name] = full_value
+
+                    # Get parameter configuration for type conversion
+                    param_config = self._get_arguments_config(
+                        self.current_function_name,
+                        self.streaming_request.tools
+                        if self.streaming_request else None)
+
+                    # Convert the parameter value to the appropriate type
+                    converted_value = self._convert_param_value(
+                        full_value, self.current_param_name, param_config,
+                        self.current_function_name)
+
+                    # Serialize the converted value
+                    serialized_value = json.dumps(converted_value,
+                                                  ensure_ascii=False)
+
+                    # Since we've been streaming the quoted version, we need to close it properly
+                    # This is complex - for now just complete the value
+                    self.in_param = False
+                    self.current_param_value = ""
+
+                    # Just close the current parameter string
+                    return DeltaMessage(tool_calls=[
+                        DeltaToolCall(
+                            index=self.current_tool_index,
+                            function=DeltaFunctionCall(
+                                arguments='"'),  # Close the string quote
+                        )
+                    ])
+                else:
+                    # Continue accumulating value
+                    value_chunk = delta_text
+
+                    # Handle first chunk after param name
+                    if not self.current_param_value and ">" in value_chunk:
+                        gt_idx = value_chunk.find(">")
+                        value_chunk = value_chunk[gt_idx + 1:]
+
+                    if not self.current_param_value and value_chunk.startswith(
+                            "\n"):
+                        value_chunk = value_chunk[1:]
+
+                    if value_chunk:
+                        # Stream the escaped delta
+                        prev_escaped = json.dumps(
+                            self.current_param_value, ensure_ascii=False
+                        )[1:-1] if self.current_param_value else ""
+                        self.current_param_value += value_chunk
+                        full_escaped = json.dumps(self.current_param_value,
+                                                  ensure_ascii=False)[1:-1]
+                        delta_escaped = full_escaped[len(prev_escaped):]
+
+                        if delta_escaped:
+                            return DeltaMessage(tool_calls=[
+                                DeltaToolCall(
+                                    index=self.current_tool_index,
+                                    function=DeltaFunctionCall(
+                                        arguments=delta_escaped),
+                                )
+                            ])
+
+        return None

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650

tokenizer_config.json

@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "is_local": true,
+  "local_files_only": false,
+  "model_max_length": 1048576,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "tool_parser_type": "qwen3_coder",
+  "unk_token": null
+}