Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,27 @@ 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
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-BF16.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F16.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F32.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ3_M.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ3_S.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ4_NL.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ4_XS.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q2_K.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q3_K_L.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q3_K_S.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_0.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_1.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_K_S.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_0.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_1.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_K_S.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q6_K.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-TQ1_0.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-TQ2_0.gguf filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-BF16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75d11ed29b7b8a7976d21c3cd6891791c4fd3b80194b50ef799ee82082550ad8
+size 53808277728

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bc9edc8d6520dd3be0f7630cd97599d1dfd1fb7048b087d0c10b1ae2643427f
+size 53808277728

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F32.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8fadcacce33789c7537e275e9f4aa8e4e9effff5444b50f2ed28c6d31f7b1e30
+size 107594983648

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ3_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7c8ca7a96a936fa2bce7c653ac7b7c75af036b0a24adebb0d814578246f4525
+size 12580870368

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ3_S.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:926c3a7a3a600910732f35131bf36a61bf51998b924e6ca6b6aaa5acfb69aa2e
+size 12419324128

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ4_NL.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:976cf75e4de43702cd50ec8a2b4be398ee8b0d782c947d84549933c10cf8e5f1
+size 15891052768

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ4_XS.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d90520006ac14e968ee2bdcc67073d642a50a26d8852cd1fc81376d835c7b86f
+size 15193913568

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q2_K.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:315187ba637f08c81b16794a32133ca7ef6c54f45e13e917aff016017358ee61
+size 10711660768

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q3_K_L.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:179e62572403b040e63e8e125888ddd485d1cbfbacaee6d1ef9039714b2f9255
+size 14344771808

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q3_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c0c2e2fbf559d62fc870ba4af1a0eb33bde592b9dfecf720b2d55f325d72a2a
+size 13301438688

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q3_K_S.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5ea1d643f27b19cd8b476f4c65eda279415f5822f16b9af0f0275896bee63c2
+size 12073949408

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a4b5b98b638ab76bc5a79bd8d453bc4f1f71d21bebcfb80a7619d776dea35c9
+size 15476865248

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_1.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37705abf793b140a8c141b88114497e4b887cb23a8def835b7909ec46dc32764
+size 17078237408

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4787c3f7241c1cbb38ca5d69104b51acdf42bf2357892f476a2d4022b7f9c004
+size 16547395808

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q4_K_S.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca6e16c8db6c8e5d6016f8c0e888dbc8d2efdc0911e47fd15ab1c8b5c6c94ec5
+size 15586310368

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5221d3ea5879717782024d29734972e0c7dec820c2c7a21d42f3f9487818a625
+size 18679609568

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_1.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78f4196934add1bf85f3c2be5139378b2384dc5adc83d33fe8c88d6bbf53c613
+size 20280981728

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34bb9278a13f92e68f7ec6c0470637e9e84a49c6aba33167e9ff5a27edcb5a01
+size 19231095008

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q5_K_S.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c013aa2c1223c76e49c4202c665a3f3432d79b8f7d9c17f7f7434e20bfb2f8d
+size 18679609568

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q6_K.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e95c10adf07a26f122f36d3e89d38ec74b465cd4ccf6a70d6dde43575da6e7ff
+size 22082525408

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6a383ae8ea725cd39568da0b6d58f3c6c521c55ecde0ae3eeb02ee0ac0e8e0a
+size 28595759328

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-TQ1_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da8ef7d5a3018e4db10e4406cebb1918b5a64a62ea0b168aa249fb6c1ba7977e
+size 6916122848

Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-TQ2_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:766bbf1215bded9624a1aec5c2f5adeeee391cf431e24ff696ce81398fe53cac
+size 8057555168

README.md

@@ -1,3 +1,139 @@
----
-license: apache-2.0
----
+---
+language:
+- en
+- zh
+license: apache-2.0
+base_model: Qwen/Qwen3.5-27B
+tags:
+- unsloth
+- qwen
+- qwen3.5
+- reasoning
+- chain-of-thought
+- Dense
+pipeline_tag: text-generation
+datasets:
+- nohurry/Opus-4.6-Reasoning-3000x-filtered
+- Jackrong/Qwen3.5-reasoning-700x
+---
+
+# 🌟 Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-GGUF
+
+> 📢 **Release Note**
+> **Build Environment Upgrades:**
+> - **Distilled and quantized on 2026.3.12**
+> - **Fine-tuning Framework**: **Unsloth 2026.3.3** 
+> - **Core Dependencies**: **Transformers 5.2.0**
+> - This model fixes the crash in the official model caused by the Jinja template not supporting the **"developer"** role. (commonly sent by modern coding agents like Claude Code and OpenCode)
+> - It does **not disable thinking mode by default**, and allowing the agent to run continuously for **over 9 minutes without interruption**.
+> - Compared to the original model, **autonomy and stability are significantly improved**.
+
+![HB8AleUaMAArNyM](https://cdn-uploads.huggingface.co/production/uploads/66309bd090589b7c65950665/GHkMJL6I383eIwK1qj80K.jpeg)
+
+
+## 💡 Model Introduction
+**Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled** is a highly capable reasoning model fine-tuned on top of the powerful Qwen3.5 architecture. The model's core directive is to leverage state-of-the-art Chain-of-Thought (CoT) distillation primarily sourced from Claude-4.6 Opus interactions. 
+
+Through Supervised Fine-Tuning (SFT) focusing specifically on structured reasoning logic, this model excels in breaking down complex user problems, planning step-by-step methodologies within strictly formatted `<think>` tags, and ultimately delivering precise, nuanced solutions. 
+
+### 🧠 Example of Learned Reasoning Scaffold（Example）
+
+The model includes targeted optimizations addressing Qwen3.5’s tendency toward excessive transitional or repetitive reasoning on simple queries. Through deep distillation and structural imitation of Claude-4.6-Opus reasoning chains, the model adopts a more efficient structured thinking pattern:  
+**“Let me analyze this request carefully: 1..2..3...”.**  
+This streamlined reasoning paradigm significantly reduces redundant cognitive loops while preserving deep analytical capacity, resulting in substantially improved inference efficiency.
+
+```text
+Let me analyze this request carefully:
+
+1. Identify the core objective of the problem.
+2. Break the task into clearly defined subcomponents.
+3. Evaluate constraints and edge cases.
+4. Formulate a step-by-step solution plan.
+5. Execute the reasoning sequentially and verify consistency.
+            .
+            .
+            .
+```
+
+## 🗺️ Training Pipeline Overview
+
+```text
+Base Model (Qwen3.5-27B)
+ │
+ ▼
+Supervised Fine-Tuning (SFT) + LoRA
+ │
+ ▼
+Final Model (Claude-4.6-Opus-Reasoning-Distilled,text-only)
+```
+
+## 📋 Stage Details
+
+🔥**Community-tested advantages** (benchmark tests by user @sudoingX on a single RTX 3090):
+
+Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled shows significant advantages in coding-agent environments such as Claude Code and OpenCode:
+
+>- **Native support for the “developer” role**, requiring no Jinja template patches or ChatML workarounds.  
+>- **Thinking mode fully preserved** (logs confirm `thinking=1`), not silently disabled, maintaining the complete chain-of-thought reasoning process.  
+>- **Greatly improved autonomy and stability** — capable of running continuously for **over 9 minutes autonomously** (with zero human intervention). It actively waits for tool responses, reads outputs, self-corrects errors, and can even automatically generate a README, whereas the base model often stalls or freezes mid-execution.  
+
+>**Hardware usage remains unchanged:**  
+>- About **16.5 GB VRAM** with **Q4_K_M** quantization  
+>- **29–35 tok/s** generation speed  
+>- **Full 262K context** with no compromises  
+
+- These improvements come from successfully distilling the **structured reasoning style of Claude 4.6 Opus**, allowing Qwopus to be truly **plug-and-play in modern local coding agents** and deliver an experience close to Opus in smoothness and usability.
+
+ **Thanks to the community for the in-depth testing and feedback!**
+
+
+### 🔹 Supervised Fine-Tuning (SFT)
+- **Objective:** To inject high-density reasoning logic and establish a strict format for problem-solving involving an internal thinking state prior to outputting the final response.
+- **Methodology:** We utilized **Unsloth** for highly efficient memory and compute optimization. A critical component of this stage is the `train_on_responses_only` strategy, masking instructions so the loss is purely calculated over the generation of the `<think>` sequences and the subsequent solutions. 
+- **Format Enforcement:** All training samples were systematically normalized so the model strictly abides by the structure `<think> {internal reasoning} </think>\n {final answer}`.
+
+### 📚 All Datasets Used
+The dataset consists of high-quality, filtered reasoning distillation data:
+
+| Dataset Name | Description / Purpose |
+|--------------|-----------------------|
+| [nohurry/Opus-4.6-Reasoning-3000x-filtered](https://huggingface.co/datasets/nohurry/Opus-4.6-Reasoning-3000x-filtered) | Provides comprehensive Claude 4.6 Opus reasoning trajectories. |
+| [TeichAI/claude-4.5-opus-high-reasoning-250x](https://huggingface.co/datasets/TeichAI/claude-4.5-opus-high-reasoning-250x) | Injecting high-intensity, structured reasoning instances. |
+| [Jackrong/Qwen3.5-reasoning-700x](https://huggingface.co/datasets/Jackrong/Qwen3.5-reasoning-700x) | Additional curated reasoning samples designed to strengthen structured step-by-step problem solving and improve reasoning diversity. |
+
+## 🌟 Core Skills & Capabilities
+1. **Modular & Structured Thinking:** Inheriting traits from Opus-level reasoning, the model demonstrates confident parsing of the prompt, establishing an outlined plan in its `<think>` block sequentially rather than exploratory "trial-and-error" self-doubt.
+
+## ⚠️ Limitations & Intended Use
+- **Hallucination Risk:** While reasoning is strong, the model remains an autoregressive LLM; external facts provided during the thinking sequence may occasionally contain hallucinations if verifying real-world events.
+- **Intended Scenario:** Best suited for offline analytical tasks, coding, math, and heavy logic-dependent prompting where the user needs to transparently follow the AI's internal logic.
+- **Preview Version Notice:** Because this model is relatively new and intentionally lightweight, the surrounding ecosystem — including inference templates, fine-tuning pipelines, routing configurations, and tooling integrations — may not yet be fully mature or standardized. As a result, users may encounter occasional bugs, compatibility inconsistencies, or integration edge cases. The current release should be considered a preview build while the broader architectural stack and supporting utilities continue to stabilize and improve.
+
+## 🙏 Acknowledgements
+Significant thanks to the [Unsloth AI](https://unsloth.ai/) team for making rapid fine-tuning of MoE and large LLM models accessible. Additionally, we acknowledge Qwen internally, and the open-source community developers producing exceptional distilled datasets (`nohurry` and `TeichAI`).
+
+## 📖 Citation
+
+If you use this model in your research or projects, please cite:
+
+```bibtex
+@misc{jackrong_qwen35_opus_distilled,
+  title        = {Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-GGUF},
+  author       = {Eugene Hauptmann},
+  year         = {2026},
+  publisher    = {Hugging Face},
+  howpublished = {\url{https://huggingface.co/eugenehp/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-GGUF}}
+}
+```
+
+Original weights:
+
+```bibtex
+@misc{jackrong_qwen35_opus_distilled,
+  title        = {Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled},
+  author       = {Jackrong},
+  year         = {2026},
+  publisher    = {Hugging Face},
+  howpublished = {\url{https://huggingface.co/Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled}}
+}
+```

SHA256SUMS

@@ -0,0 +1,13 @@
+75d11ed29b7b8a7976d21c3cd6891791c4fd3b80194b50ef799ee82082550ad8  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-BF16.gguf
+75d11ed29b7b8a7976d21c3cd6891791c4fd3b80194b50ef799ee82082550ad8  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-BF16.gguf
+75d11ed29b7b8a7976d21c3cd6891791c4fd3b80194b50ef799ee82082550ad8  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-BF16.gguf
+75d11ed29b7b8a7976d21c3cd6891791c4fd3b80194b50ef799ee82082550ad8  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-BF16.gguf
+1bc9edc8d6520dd3be0f7630cd97599d1dfd1fb7048b087d0c10b1ae2643427f  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F16.gguf
+8fadcacce33789c7537e275e9f4aa8e4e9effff5444b50f2ed28c6d31f7b1e30  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F32.gguf
+f7c8ca7a96a936fa2bce7c653ac7b7c75af036b0a24adebb0d814578246f4525  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ3_M.gguf
+926c3a7a3a600910732f35131bf36a61bf51998b924e6ca6b6aaa5acfb69aa2e  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ3_S.gguf
+976cf75e4de43702cd50ec8a2b4be398ee8b0d782c947d84549933c10cf8e5f1  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ4_NL.gguf
+d90520006ac14e968ee2bdcc67073d642a50a26d8852cd1fc81376d835c7b86f  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-IQ4_XS.gguf
+75d11ed29b7b8a7976d21c3cd6891791c4fd3b80194b50ef799ee82082550ad8  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-BF16.gguf
+1bc9edc8d6520dd3be0f7630cd97599d1dfd1fb7048b087d0c10b1ae2643427f  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F16.gguf
+8fadcacce33789c7537e275e9f4aa8e4e9effff5444b50f2ed28c6d31f7b1e30  Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-F32.gguf

chat_template.jinja

@@ -0,0 +1,88 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant
+<think>
+' }}
+{%- endif %}

config.json

@@ -0,0 +1,146 @@
+{
+    "architectures": [
+        "Qwen3_5ForConditionalGeneration"
+    ],
+    "bos_token_id": null,
+    "torch_dtype": "bfloat16",
+    "eos_token_id": 248046,
+    "image_token_id": 248056,
+    "model_name": "qwen/Qwen3.5-27B",
+    "model_type": "qwen3_5",
+    "pad_token_id": 248044,
+    "text_config": {
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "attn_output_gate": true,
+        "bos_token_id": null,
+        "torch_dtype": "bfloat16",
+        "eos_token_id": 248044,
+        "full_attention_interval": 4,
+        "head_dim": 256,
+        "hidden_act": "silu",
+        "hidden_size": 5120,
+        "initializer_range": 0.02,
+        "intermediate_size": 17408,
+        "layer_types": [
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention"
+        ],
+        "linear_conv_kernel_dim": 4,
+        "linear_key_head_dim": 128,
+        "linear_num_key_heads": 16,
+        "linear_num_value_heads": 48,
+        "linear_value_head_dim": 128,
+        "mamba_ssm_dtype": "float32",
+        "max_position_embeddings": 262144,
+        "mlp_only_layers": [],
+        "model_type": "qwen3_5_text",
+        "mtp_num_hidden_layers": 1,
+        "mtp_use_dedicated_embeddings": false,
+        "num_attention_heads": 24,
+        "num_hidden_layers": 64,
+        "num_key_value_heads": 4,
+        "pad_token_id": null,
+        "partial_rotary_factor": 0.25,
+        "rms_norm_eps": 1e-06,
+        "rope_parameters": {
+            "mrope_interleaved": true,
+            "mrope_section": [
+                11,
+                11,
+                10
+            ],
+            "partial_rotary_factor": 0.25,
+            "rope_theta": 10000000,
+            "rope_type": "default"
+        },
+        "tie_word_embeddings": false,
+        "use_cache": true,
+        "vocab_size": 248320
+    },
+    "tie_word_embeddings": false,
+    "unsloth_version": "2026.3.3",
+    "use_cache": false,
+    "video_token_id": 248057,
+    "vision_config": {
+        "deepstack_visual_indexes": [],
+        "depth": 27,
+        "torch_dtype": "bfloat16",
+        "hidden_act": "gelu_pytorch_tanh",
+        "hidden_size": 1152,
+        "in_channels": 3,
+        "initializer_range": 0.02,
+        "intermediate_size": 4304,
+        "model_type": "qwen3_5",
+        "num_heads": 16,
+        "num_position_embeddings": 2304,
+        "out_hidden_size": 5120,
+        "patch_size": 16,
+        "spatial_merge_size": 2,
+        "temporal_patch_size": 2
+    },
+    "vision_end_token_id": 248054,
+    "vision_start_token_id": 248053
+}

processor_config.json

@@ -0,0 +1,63 @@
+{
+  "image_processor": {
+    "data_format": "channels_first",
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "Qwen2VLImageProcessorFast",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "merge_size": 2,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "longest_edge": 16777216,
+      "shortest_edge": 65536
+    },
+    "temporal_patch_size": 2
+  },
+  "processor_class": "Qwen3VLProcessor",
+  "video_processor": {
+    "data_format": "channels_first",
+    "default_to_square": true,
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "fps": 2,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "max_frames": 768,
+    "merge_size": 2,
+    "min_frames": 4,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "size": {
+      "longest_edge": 25165824,
+      "shortest_edge": 4096
+    },
+    "temporal_patch_size": 2,
+    "video_processor_type": "Qwen3VLVideoProcessor"
+  }
+}

scripts/convert_to_gguf.sh

@@ -0,0 +1,269 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+#=============================================================================
+# Convert Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled to GGUF
+#
+# Produces ALL quantization variants from a single F16 GGUF base.
+#
+# Prerequisites:
+#   - Python 3.10+
+#   - pip packages: numpy torch transformers sentencepiece safetensors gguf protobuf
+#   - llama.cpp repo cloned (for convert_hf_to_gguf.py and llama-quantize)
+#
+# Usage:
+#   ./convert_to_gguf.sh                        # all quants, default llama.cpp path
+#   ./convert_to_gguf.sh /path/to/llama.cpp     # all quants, custom llama.cpp path
+#=============================================================================
+
+MODEL_REPO="Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled"
+MODEL_NAME="Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled"
+LLAMA_CPP_DIR="${1:-$HOME/Desktop/llama.cpp}"
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+OUTPUT_DIR="${SCRIPT_DIR}/gguf_output"
+
+# ─── Every quantization type supported by llama-quantize ─────────────────────
+# Sorted roughly by bits-per-weight (smallest → largest)
+QUANT_TYPES=(
+    # ~1-2 bpw — extreme compression
+    IQ1_S        # 1.56 bpw
+    IQ1_M        # 1.75 bpw
+    TQ1_0        # 1.69 bpw ternary
+    TQ2_0        # 2.06 bpw ternary
+    IQ2_XXS      # 2.06 bpw
+    IQ2_XS       # 2.31 bpw
+    IQ2_S        # 2.5  bpw
+    IQ2_M        # 2.7  bpw
+
+    # ~3 bpw
+    Q2_K         # 2.96G @ 8B, +3.52 ppl
+    Q2_K_S       # 2.96G @ 8B, +3.18 ppl
+    IQ3_XXS      # 3.06 bpw
+    IQ3_XS       # 3.3  bpw
+    IQ3_S        # 3.44 bpw
+    IQ3_M        # 3.66 bpw
+    Q3_K_S       # 3.41G @ 8B, +1.63 ppl
+    Q3_K_M       # 3.74G @ 8B, +0.66 ppl
+    Q3_K_L       # 4.03G @ 8B, +0.56 ppl
+
+    # ~4 bpw — sweet spot
+    IQ4_XS       # 4.25 bpw
+    IQ4_NL       # 4.50 bpw
+    Q4_0         # 4.34G @ 8B, +0.47 ppl
+    Q4_1         # 4.78G @ 8B, +0.45 ppl
+    Q4_K_S       # 4.37G @ 8B, +0.27 ppl
+    Q4_K_M       # 4.58G @ 8B, +0.18 ppl  ← popular default
+
+    # ~5 bpw
+    Q5_0         # 5.21G @ 8B, +0.13 ppl
+    Q5_1         # 5.65G @ 8B, +0.11 ppl
+    Q5_K_S       # 5.21G @ 8B, +0.10 ppl
+    Q5_K_M       # 5.33G @ 8B, +0.06 ppl
+
+    # ~6-8 bpw — high fidelity
+    Q6_K         # 6.14G @ 8B, +0.02 ppl
+    Q8_0         # 7.96G @ 8B, +0.003 ppl
+
+    # Full precision (format conversion only)
+    BF16         # 14G @ 7B
+    F16          # 14G @ 7B
+    F32          # 26G @ 7B
+)
+
+echo "============================================="
+echo " HuggingFace → GGUF Conversion Script"
+echo "============================================="
+echo " Model:       $MODEL_REPO"
+echo " llama.cpp:   $LLAMA_CPP_DIR"
+echo " Output dir:  $OUTPUT_DIR"
+echo " Quant types: ${#QUANT_TYPES[@]} variants"
+echo "============================================="
+
+#-----------------------------------------------------------------------------
+# Step 0: Validate llama.cpp directory
+#-----------------------------------------------------------------------------
+if [ ! -f "$LLAMA_CPP_DIR/convert_hf_to_gguf.py" ]; then
+    echo "❌ convert_hf_to_gguf.py not found in $LLAMA_CPP_DIR"
+    echo "   Clone llama.cpp first:  git clone https://github.com/ggml-org/llama.cpp.git"
+    exit 1
+fi
+
+#-----------------------------------------------------------------------------
+# Step 1: Resolve model from HF cache, local dir, or download
+#-----------------------------------------------------------------------------
+echo ""
+echo "▶ Step 1: Resolving model files..."
+
+HF_CACHE_DIR="$HOME/.cache/huggingface/hub"
+MODEL_CACHE="$HF_CACHE_DIR/models--$(echo "$MODEL_REPO" | tr '/' '--')"
+MODEL_DIR=""
+
+# Helper: check if a directory has config.json + at least one .safetensors
+has_model_files() {
+    local dir="$1"
+    [ -f "$dir/config.json" ] && \
+    [ "$(find "$dir" -maxdepth 1 -name '*.safetensors' 2>/dev/null | head -1)" != "" ]
+}
+
+# 1) Check HF cache snapshots
+if [ -d "$MODEL_CACHE/snapshots" ]; then
+    SNAPSHOT=$(ls -t "$MODEL_CACHE/snapshots" 2>/dev/null | head -1)
+    if [ -n "$SNAPSHOT" ]; then
+        CANDIDATE="$MODEL_CACHE/snapshots/$SNAPSHOT"
+        if has_model_files "$CANDIDATE"; then
+            MODEL_DIR="$CANDIDATE"
+            echo "  Found complete model in HF cache: $MODEL_DIR"
+        else
+            echo "  HF cache exists but is incomplete (missing safetensors or config.json)"
+        fi
+    fi
+fi
+
+# 2) Check local model_hf directory (from a previous download)
+if [ -z "$MODEL_DIR" ] && has_model_files "${SCRIPT_DIR}/model_hf"; then
+    MODEL_DIR="${SCRIPT_DIR}/model_hf"
+    echo "  Found complete model in local dir: $MODEL_DIR"
+fi
+
+# 3) Nothing found — download everything
+if [ -z "$MODEL_DIR" ]; then
+    MODEL_DIR="${SCRIPT_DIR}/model_hf"
+    echo "  ⬇  No complete model found locally. Downloading from HuggingFace..."
+    echo "     Repo: $MODEL_REPO"
+    echo "     Dest: $MODEL_DIR"
+    echo ""
+    python3 -c "
+from huggingface_hub import snapshot_download
+path = snapshot_download(
+    '$MODEL_REPO',
+    local_dir='$MODEL_DIR',
+    resume_download=True,
+)
+print(f'  ✅ Downloaded to: {path}')
+"
+    echo ""
+
+    # Verify the download actually worked
+    if ! has_model_files "$MODEL_DIR"; then
+        echo "❌ Download finished but model files are still missing."
+        echo "   Expected config.json and *.safetensors in: $MODEL_DIR"
+        echo ""
+        echo "   Contents:"
+        ls -la "$MODEL_DIR" 2>/dev/null || echo "   (directory does not exist)"
+        exit 1
+    fi
+fi
+
+SAFETENSOR_COUNT=$(find "$MODEL_DIR" -maxdepth 1 -name "*.safetensors" 2>/dev/null | wc -l)
+TOTAL_SIZE=$(du -sh "$MODEL_DIR"/*.safetensors 2>/dev/null | tail -1 | cut -f1)
+echo "  Found config.json and $SAFETENSOR_COUNT safetensor shard(s) (~${TOTAL_SIZE:-?} total)"
+
+#-----------------------------------------------------------------------------
+# Step 2: Convert to F16 GGUF (base for all quantizations)
+#-----------------------------------------------------------------------------
+echo ""
+echo "▶ Step 2: Converting safetensors → GGUF (F16 base)..."
+
+mkdir -p "$OUTPUT_DIR"
+F16_GGUF="$OUTPUT_DIR/${MODEL_NAME}-F16.gguf"
+
+if [ -f "$F16_GGUF" ]; then
+    echo "  F16 GGUF already exists, skipping conversion."
+else
+    python3 "$LLAMA_CPP_DIR/convert_hf_to_gguf.py" \
+        "$MODEL_DIR" \
+        --outfile "$F16_GGUF" \
+        --outtype f16
+
+    echo "  ✅ Created: $F16_GGUF"
+fi
+
+F16_SIZE=$(du -h "$F16_GGUF" | cut -f1)
+echo "  F16 size: $F16_SIZE"
+
+#-----------------------------------------------------------------------------
+# Step 3: Build llama-quantize if needed
+#-----------------------------------------------------------------------------
+echo ""
+echo "▶ Step 3: Ensuring llama-quantize is built..."
+
+QUANTIZE_BIN="$LLAMA_CPP_DIR/build/bin/llama-quantize"
+if [ ! -f "$QUANTIZE_BIN" ]; then
+    echo "  Building llama-quantize..."
+    pushd "$LLAMA_CPP_DIR" > /dev/null
+    cmake -B build -DCMAKE_BUILD_TYPE=Release 2>&1 | tail -3
+    cmake --build build --target llama-quantize -j"$(nproc)" 2>&1 | tail -5
+    popd > /dev/null
+
+    if [ ! -f "$QUANTIZE_BIN" ]; then
+        echo "❌ Failed to build llama-quantize"
+        exit 1
+    fi
+fi
+echo "  ✅ llama-quantize ready: $QUANTIZE_BIN"
+
+#-----------------------------------------------------------------------------
+# Step 4: Quantize ALL variants
+#-----------------------------------------------------------------------------
+echo ""
+echo "▶ Step 4: Quantizing ${#QUANT_TYPES[@]} variants..."
+echo ""
+
+SUCCEEDED=0
+FAILED=0
+SKIPPED=0
+FAILED_LIST=()
+
+for QTYPE in "${QUANT_TYPES[@]}"; do
+    QUANT_GGUF="$OUTPUT_DIR/${MODEL_NAME}-${QTYPE}.gguf"
+
+    # Skip F16 since that's our base file
+    if [ "$QTYPE" = "F16" ]; then
+        echo "  [SKIP] $QTYPE — already created as base"
+        SKIPPED=$((SKIPPED + 1))
+        continue
+    fi
+
+    # Skip if already exists
+    if [ -f "$QUANT_GGUF" ]; then
+        SIZE=$(du -h "$QUANT_GGUF" | cut -f1)
+        echo "  [SKIP] $QTYPE — already exists ($SIZE)"
+        SKIPPED=$((SKIPPED + 1))
+        continue
+    fi
+
+    echo -n "  [QUANT] $QTYPE ... "
+    if "$QUANTIZE_BIN" "$F16_GGUF" "$QUANT_GGUF" "$QTYPE" > "$OUTPUT_DIR/.quantize_${QTYPE}.log" 2>&1; then
+        SIZE=$(du -h "$QUANT_GGUF" | cut -f1)
+        echo "✅ ($SIZE)"
+        SUCCEEDED=$((SUCCEEDED + 1))
+    else
+        echo "❌ FAILED (see $OUTPUT_DIR/.quantize_${QTYPE}.log)"
+        FAILED=$((FAILED + 1))
+        FAILED_LIST+=("$QTYPE")
+    fi
+done
+
+#-----------------------------------------------------------------------------
+# Summary
+#-----------------------------------------------------------------------------
+echo ""
+echo "============================================="
+echo " ✅ Conversion complete!"
+echo "============================================="
+echo ""
+echo " Results: $SUCCEEDED succeeded, $SKIPPED skipped, $FAILED failed"
+if [ ${#FAILED_LIST[@]} -gt 0 ]; then
+    echo " Failed:  ${FAILED_LIST[*]}"
+fi
+echo ""
+echo " Output directory: $OUTPUT_DIR"
+echo ""
+echo " Files:"
+echo " ─────"
+ls -lhS "$OUTPUT_DIR"/*.gguf 2>/dev/null | awk '{printf "   %-12s %s\n", $5, $NF}'
+echo ""
+echo " Test with llama.cpp:"
+echo "   $LLAMA_CPP_DIR/build/bin/llama-cli -m $OUTPUT_DIR/${MODEL_NAME}-Q4_K_M.gguf -p 'Hello!' -n 128"
+echo "============================================="
+

scripts/verify_gguf.sh

@@ -0,0 +1,221 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+export PYTHONPATH=$HOME/Desktop/llama.cpp/gguf-py
+
+# ──────────────────────────────────────────────────────────────
+# Verify GGUF files
+# - Metadata
+# - Tensor inventory (optional, skips if missing)
+# - SHA256 hash
+# - Smoke-test inference
+# - Perplexity (optional)
+# Usage:
+#   ./verify_gguf.sh [--perplexity] [file1.gguf file2.gguf ...]
+# ──────────────────────────────────────────────────────────────
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+LLAMA_CPP_DIR="${LLAMA_CPP_DIR:-$HOME/Desktop/llama.cpp}"
+OUTPUT_DIR="${SCRIPT_DIR}/gguf_output"
+GGUF_PY_DIR="$LLAMA_CPP_DIR/gguf-py"
+RUN_PERPLEXITY=false
+TARGETS=()
+
+# Add gguf-py to PYTHONPATH
+export PYTHONPATH="$GGUF_PY_DIR:$PYTHONPATH"
+
+# Detect number of CPU cores
+if command -v nproc >/dev/null 2>&1; then
+    NPROC=$(nproc)
+else
+    NPROC=$(sysctl -n hw.ncpu)
+fi
+
+# ─── Parse arguments ─────────────────────────────────────────────
+while [[ $# -gt 0 ]]; do
+    case "$1" in
+        --perplexity|--ppl|--all)
+            RUN_PERPLEXITY=true
+            shift
+            ;;
+        --llama-cpp)
+            LLAMA_CPP_DIR="$2"
+            GGUF_PY_DIR="$LLAMA_CPP_DIR/gguf-py"
+            shift 2
+            ;;
+        *)
+            TARGETS+=("$1")
+            shift
+            ;;
+    esac
+done
+
+# ─── If no targets, find all GGUFs in output dir ───────────────
+if [ ${#TARGETS[@]} -eq 0 ]; then
+    while IFS= read -r f; do
+        TARGETS+=("$f")
+    done < <(find "$OUTPUT_DIR" -name "*.gguf" -type f | sort)
+fi
+
+if [ ${#TARGETS[@]} -eq 0 ]; then
+    echo "❌ No GGUF files found in $OUTPUT_DIR"
+    exit 1
+fi
+
+# ─── Build llama-cli / llama-perplexity if missing ─────────────
+CLI_BIN="$LLAMA_CPP_DIR/build/bin/llama-cli"
+PPL_BIN="$LLAMA_CPP_DIR/build/bin/llama-perplexity"
+
+build_target() {
+    local target="$1"
+    local bin="$LLAMA_CPP_DIR/build/bin/$target"
+    if [ ! -f "$bin" ]; then
+        echo "  Building $target..."
+        pushd "$LLAMA_CPP_DIR" > /dev/null
+        cmake -B build -DCMAKE_BUILD_TYPE=Release 2>&1 | tail -1
+        cmake --build build --target "$target" -j"$NPROC" 2>&1 | tail -3
+        popd > /dev/null
+    fi
+}
+
+build_target "llama-cli"
+if $RUN_PERPLEXITY; then
+    build_target "llama-perplexity"
+fi
+
+# ─── Begin verification ────────────────────────────────────────
+PASS=0
+FAIL=0
+
+echo "============================================="
+echo " GGUF Verification"
+echo "============================================="
+echo " Files to check: ${#TARGETS[@]}"
+echo " Perplexity:     $RUN_PERPLEXITY"
+echo "============================================="
+echo ""
+
+for GGUF_FILE in "${TARGETS[@]}"; do
+    BASENAME=$(basename "$GGUF_FILE")
+    echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
+    echo "  $BASENAME"
+    echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
+
+    # Skip empty or missing files
+    if [ ! -s "$GGUF_FILE" ]; then
+        echo "  ⚠️  Skipping empty or missing file"
+        FAIL=$((FAIL+1))
+        continue
+    fi
+
+    FILE_SIZE=$(du -h "$GGUF_FILE" | cut -f1)
+    echo "  Size: $FILE_SIZE"
+    echo ""
+
+    FILE_OK=true
+
+    # ── Check 1: Metadata ─────────────────────────────
+    echo "  ┌─ [1/4] Metadata"
+    if METADATA=$(python3 "$GGUF_PY_DIR/gguf/scripts/gguf_dump.py" "$GGUF_FILE" --no-tensors 2>&1); then
+        ARCH=$(echo "$METADATA" | sed -n 's/.*general\.architecture *= *\([^ ]*\).*/\1/p' | head -1)
+        QTYPE=$(echo "$METADATA" | sed -n 's/.*general\.file_type *= *\(.*\)/\1/p' | head -1)
+        CTX=$(echo "$METADATA" | sed -n 's/.*context_length *= *\([^ ]*\).*/\1/p' | head -1)
+        NLAYER=$(echo "$METADATA" | sed -n 's/.*block_count *= *\([^ ]*\).*/\1/p' | head -1)
+        VOCAB=$(echo "$METADATA" | sed -n 's/.*tokens.*length *= *\([^ ]*\).*/\1/p' | head -1)
+        echo "  │   Architecture:    ${ARCH:-?}"
+        echo "  │   File type:       ${QTYPE:-?}"
+        echo "  │   Context length:  ${CTX:-?}"
+        echo "  │   Layers:          ${NLAYER:-?}"
+        echo "  │   Vocab size:      ${VOCAB:-?}"
+        echo "  │   ✅ Metadata OK"
+    else
+        echo "  │   ❌ Failed to read metadata"
+        echo "  │   $METADATA" | head -5
+        FILE_OK=false
+        # Skip tensor check if metadata fails
+        SKIP_TENSORS=true
+    fi
+    echo "  │"
+
+    # ── Check 2: Tensor inventory ─────────────────────
+    echo "  ├─ [2/4] Tensor inventory"
+    if [ "${SKIP_TENSORS:-false}" = true ]; then
+        echo "  │   ⚠️  Skipping tensors due to missing metadata"
+    else
+        if TENSOR_INFO=$(python3 "$GGUF_PY_DIR/gguf/scripts/gguf_dump.py" "$GGUF_FILE" --json 2>/dev/null); then
+            if [ -z "$TENSOR_INFO" ]; then
+                echo "  │   ⚠️  Tensor info empty, skipping"
+            else
+                TENSOR_COUNT=$(echo "$TENSOR_INFO" | python3 -c "
+import sys, json
+data = json.load(sys.stdin)
+tensors = data.get('tensors', [])
+print(len(tensors))
+" 2>/dev/null || echo "?")
+
+                echo "  │   Total tensors: ${TENSOR_COUNT:-?}"
+                echo "  │   ✅ Tensors OK"
+            fi
+        else
+            echo "  │   ⚠️  Failed to parse tensor JSON, skipping"
+        fi
+    fi
+    echo "  │"
+
+    # ── Check 3: SHA256 hash ──────────────────────────
+    echo "  ├─ [3/4] SHA256 hash"
+    if command -v shasum >/dev/null 2>&1; then
+        HASH=$(shasum -a 256 "$GGUF_FILE" | cut -d' ' -f1)
+    else
+        HASH=$(sha256sum "$GGUF_FILE" | cut -d' ' -f1)
+    fi
+    echo "  │   $HASH"
+    echo "  │   ✅ Hash computed"
+    echo "$HASH  $BASENAME" >> "$OUTPUT_DIR/SHA256SUMS"
+    echo "  │"
+
+    # ── Check 4: Smoke-test inference ───────────────
+    echo "  └─ [4/4] Smoke-test inference"
+    if [ ! -s "$GGUF_FILE" ]; then
+        echo "  ⚠️  File empty or missing weights, skipping inference"
+        FILE_OK=false
+    else
+        SMOKE_OUTPUT=$("$CLI_BIN" -m "$GGUF_FILE" -p "The capital of France is" -n 20 -r "" --no-display-prompt --seed 42 --temp 0.0 --threads 4 --log-disable 2>&1) || FILE_OK=false
+        TOKEN_COUNT=$(echo "$SMOKE_OUTPUT" | wc -w)
+        PREVIEW=$(echo "$SMOKE_OUTPUT" | head -3 | sed 's/^/  │   > /')
+        echo "$PREVIEW"
+        if [ "$TOKEN_COUNT" -gt 0 ]; then
+            echo "  │   ✅ Generated $TOKEN_COUNT words"
+        else
+            echo "  │   ⚠️  Empty output"
+        fi
+    fi
+    echo ""
+
+    # ── Summary per file ─────────────────────────────
+    if $FILE_OK; then
+        echo "  ✅ $BASENAME — ALL CHECKS PASSED"
+        PASS=$((PASS + 1))
+    else
+        echo "  ❌ $BASENAME — SOME CHECKS FAILED"
+        FAIL=$((FAIL + 1))
+    fi
+    echo ""
+done
+
+# ─── Final summary ─────────────────────────────────────────────
+echo "============================================="
+echo " Verification Summary"
+echo "============================================="
+echo " Passed: $PASS"
+echo " Failed: $FAIL"
+echo " Total:  ${#TARGETS[@]}"
+if [ -f "$OUTPUT_DIR/SHA256SUMS" ]; then
+    sort -u -o "$OUTPUT_DIR/SHA256SUMS" "$OUTPUT_DIR/SHA256SUMS"
+    echo ""
+    echo " SHA256 checksums saved to:"
+    echo "   $OUTPUT_DIR/SHA256SUMS"
+fi
+echo "============================================="
+
+exit $FAIL

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87a7830d63fcf43bf241c3c5242e96e62dd3fdc29224ca26fed8ea333db72de4
+size 19989343

tokenizer_config.json

@@ -0,0 +1,34 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "processor_class": "Qwen3VLProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>",
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0].role == 'system' %}\n        {{- messages[0].content + '\\n\\n' }}\n    {%- endif %}\n    {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0].content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}\n        {%- set ns.multi_step_tool = false %}\n        {%- set ns.last_query_index = index %}\n    {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n    {%- if message.content is string %}\n        {%- set content = message.content %}\n    {%- else %}\n        {%- set content = '' %}\n    {%- endif %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n        {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- if loop.index0 > ns.last_query_index %}\n            {%- if loop.last or (not loop.last and reasoning_content) %}\n                {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content.strip('\\n') + '\\n</think>\\n\\n' + content.lstrip('\\n') }}\n            {%- else %}\n                {{- '<|im_start|>' + message.role + '\\n' + content }}\n            {%- endif %}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if (loop.first and content) or (not loop.first) %}\n                    {{- '\\n' }}\n                {%- endif %}\n                {%- if tool_call.function %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {{- '<tool_call>\\n{\"name\": \"' }}\n                {{- tool_call.name }}\n                {{- '\", \"arguments\": ' }}\n                {%- if tool_call.arguments is string %}\n                    {{- tool_call.arguments }}\n                {%- else %}\n                    {{- tool_call.arguments | tojson }}\n                {%- endif %}\n                {{- '}\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\n<think>\n' }}\n{%- endif %}"
+}