Update README.md

README.md

@@ -1,207 +1,282 @@
 ---
 license: apache-2.0
 language:
-- en
-- es
-base_model:
-- Qwen/Qwen3.8-27B
+  - en
+  - es
+  - zh
+base_model: Qwen/Qwen3.8-27B
 pipeline_tag: text-generation
 tags:
-- >-
-  gguf, qwen, qwen3.8, 27b, q4_k_m, mtp, nextn, llama.cpp, ollama, quantized,
-  conversational
+  - gguf
+  - qwen
+  - qwen3.8
+  - 27b
+  - q4_k_m
+  - mtp
+  - nextn
+  - llama.cpp
+  - ollama
+  - quantized
+  - conversational
 ---
-# Model Card for Model ID
 
-<!-- Provide a quick summary of what the model is/does. -->
+# Qwen3.8-27B MTP GGUF — Q4_K_M
 
-This modelcard aims to be a base template for new models. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/modelcard_template.md?plain=1).
-
-## Model Details
-
-### Model Description
-
-<!-- Provide a longer summary of what this model is. -->
+Community **GGUF Q4_K_M quantization of `Qwen/Qwen3.8-27B` with the original MTP / NextN tensors preserved**.
 
+This repository contains a format conversion and quantization of the original Qwen3.8-27B checkpoint.
 
+**No fine-tuning or additional training has been performed.**
 
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
+## Model Details
 
-### Model Sources [optional]
+| Property     | Value              |
+| ------------ | ------------------ |
+| Base model   | `Qwen/Qwen3.8-27B` |
+| Parameters   | 27B                |
+| Quantization | `Q4_K_M`           |
+| Format       | GGUF               |
+| MTP / NextN  | Preserved          |
+| File size    | ~16.8 GB           |
+| Conversion   | `llama.cpp`        |
+| License      | Apache-2.0         |
 
-<!-- Provide the basic links for the model. -->
+### Available File
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+```text
+Qwen3.8-27B-Q4_K_M-MTP.gguf
+```
 
-## Uses
+## MTP / NextN
 
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+The original Qwen3.8-27B checkpoint contains Multi-Token Prediction / NextN tensors.
 
-### Direct Use
+During conversion:
 
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+```text
+MTP enabled: 866 tensors
+NextN disabled: 851 tensors
+Difference: 15 MTP tensors
+```
 
-[More Information Needed]
+The 15 additional `mtp.*` tensors were intentionally preserved in this GGUF.
 
-### Downstream Use [optional]
+The current `llama.cpp` converter recognized the MTP export path and correctly mapped the NextN tensors into the additional model block.
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+## llama.cpp Validation
 
-[More Information Needed]
+The resulting Q4_K_M GGUF was successfully loaded and executed with `llama.cpp` using:
 
-### Out-of-Scope Use
+```bash
+llama-cli \
+  -m Qwen3.8-27B-Q4_K_M-MTP.gguf \
+  --spec-type draft-mtp
+```
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+The runtime successfully:
 
-[More Information Needed]
+* loaded the GGUF;
+* recognized the model architecture;
+* recognized the embedded MTP / NextN tensors;
+* constructed the main and MTP graphs;
+* processed the prompt;
+* generated output through the MTP-compatible runtime path.
 
-## Bias, Risks, and Limitations
+## Ollama
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+The GGUF has also been successfully imported and executed with **Ollama 0.32.9**.
 
-[More Information Needed]
+Minimal `Modelfile`:
 
-### Recommendations
+```text
+FROM ./Qwen3.8-27B-Q4_K_M-MTP.gguf
+```
 
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+Create:
 
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+```bash
+ollama create qwen3.8:27b-mtp-q4_K_M -f Modelfile
+```
 
-## How to Get Started with the Model
+Run:
 
-Use the code below to get started with the model.
+```bash
+ollama run qwen3.8:27b-mtp-q4_K_M
+```
 
-[More Information Needed]
+### MTP Configuration
 
-## Training Details
+For runtimes that expose MTP speculative decoding, the local TERATHOX configuration uses:
 
-### Training Data
+```text
+draft_num_predict = 4
+```
 
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+Note that **loading a GGUF containing MTP tensors does not by itself guarantee that a runtime is actively using speculative MTP decoding**.
 
-[More Information Needed]
+Users should verify MTP support and configuration for their specific runtime version.
 
-### Training Procedure
+## Local TERATHOX Deployment
 
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+This quantization has been tested locally under the alias:
 
-#### Preprocessing [optional]
+```text
+Terathox-Coder:Nova
+```
 
-[More Information Needed]
+### Hardware
 
+```text
+NVIDIA GeForce RTX 5080  16 GB
+NVIDIA GeForce RTX 4070  12 GB
+NVIDIA GeForce RTX 4070  12 GB
+```
 
-#### Training Hyperparameters
+Three GPUs were used for the local validation.
 
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+### Ollama Runtime Configuration
 
-#### Speeds, Sizes, Times [optional]
+```text
+Context:                  204800
+OLLAMA_FLASH_ATTENTION:   1
+OLLAMA_VULKAN:            false
+OLLAMA_KV_CACHE_TYPE:     q4_0
+OLLAMA_SCHED_SPREAD:      false
+OLLAMA_GPU_OVERHEAD:      0
+OLLAMA_NUM_PARALLEL:      1
+OLLAMA_MAX_LOADED_MODELS: 1
+OLLAMA_KEEP_ALIVE:        -1
+draft_num_predict:        4
+```
 
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+Observed status:
 
-[More Information Needed]
+```text
+NAME                   SIZE     PROCESSOR    CONTEXT
+Terathox-Coder:Nova    24 GB    100% GPU     204800
+```
 
-## Evaluation
+## Local Performance
 
-<!-- This section describes the evaluation protocols and provides the results. -->
+Observed interactive generation performance:
 
-### Testing Data, Factors & Metrics
+| Run |   Eval rate |
+| --- | ----------: |
+| 1   | 49.90 tok/s |
+| 2   | 49.52 tok/s |
+| 3   | 54.18 tok/s |
+| 4   | 49.32 tok/s |
 
-#### Testing Data
+Typical observed generation range:
 
-<!-- This should link to a Dataset Card if possible. -->
+```text
+~49–54 tokens/s
+```
 
-[More Information Needed]
+Prompt evaluation varied depending on conversation state and cached context, reaching values from approximately:
 
-#### Factors
+```text
+51 tok/s → 263 tok/s
+```
 
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+These are **local hardware measurements and not standardized model benchmarks**.
 
-[More Information Needed]
+Performance depends on hardware, context size, GPU offload, KV-cache configuration, runtime version and MTP implementation.
 
-#### Metrics
+## Intended Use
 
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+This GGUF is intended for:
 
-[More Information Needed]
+* local text generation;
+* coding and software engineering;
+* agentic coding workflows;
+* technical reasoning;
+* long-context workloads;
+* experimentation with MTP / NextN speculative decoding;
+* local inference with `llama.cpp` or compatible GGUF runtimes.
 
-### Results
+## Limitations
 
-[More Information Needed]
+This is a quantized derivative of the original model.
 
-#### Summary
+`Q4_K_M` significantly reduces memory requirements but may introduce some quality degradation compared with the original BF16 checkpoint.
 
+The base model may also produce inaccurate, biased or hallucinated information. Outputs should be independently verified for high-impact or safety-critical use cases.
 
+## Vision / Multimodal Support
 
-## Model Examination [optional]
+The original Qwen3.8-27B model includes multimodal capabilities.
 
-<!-- Relevant interpretability work for the model goes here -->
+**This repository currently provides the GGUF language-model artifact only.**
 
-[More Information Needed]
+No independently validated multimodal projector (`mmproj`) is currently included in this repository.
 
-## Environmental Impact
+Therefore this release should currently be considered **text-oriented unless an appropriate multimodal projector is added and validated**.
 
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+## Training
 
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+No training or fine-tuning was performed for this repository.
 
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
+The original weights come from:
 
-## Technical Specifications [optional]
+```text
+Qwen/Qwen3.8-27B
+```
 
-### Model Architecture and Objective
+This repository only performs:
 
-[More Information Needed]
+```text
+Original checkpoint
+        ↓
+GGUF BF16 with MTP preserved
+        ↓
+Q4_K_M quantization
+        ↓
+Qwen3.8-27B-Q4_K_M-MTP.gguf
+```
 
-### Compute Infrastructure
+## Datasets
 
-[More Information Needed]
+No additional dataset was used.
 
-#### Hardware
+This repository does not contain a fine-tuned model.
 
-[More Information Needed]
+## Evaluation
 
-#### Software
+No standardized quality benchmark was performed specifically on this quantization at the time of publication.
 
-[More Information Needed]
+The performance results above measure **local inference throughput only** and should not be interpreted as accuracy or capability benchmarks.
 
-## Citation [optional]
+For official capability benchmarks, refer to the original `Qwen/Qwen3.8-27B` model card.
 
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+## Attribution
 
-**BibTeX:**
+Original foundation model developed by the **Qwen Team**.
 
-[More Information Needed]
+Base model:
 
-**APA:**
+```text
+Qwen/Qwen3.8-27B
+```
 
-[More Information Needed]
+GGUF conversion and Q4_K_M quantization:
 
-## Glossary [optional]
+```text
+Terathox-Coder
+```
 
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+The original MTP / NextN tensors were preserved during conversion.
 
-[More Information Needed]
+**TERATHOX does not claim authorship or training of the original Qwen foundation model.**
 
-## More Information [optional]
+## License
 
-[More Information Needed]
+This repository follows the **Apache License 2.0** of the base model.
 
-## Model Card Authors [optional]
+Please review the original Qwen3.8-27B repository and license for additional information.
 
-[More Information Needed]
+## Disclaimer
 
-## Model Card Contact
+This is a **community conversion** and is not an official Qwen release.
 
-[More Information Needed]
+Compatibility, performance and MTP behavior may vary between versions of `llama.cpp`, Ollama and other GGUF runtimes.