arcee-ai
/

Virtuoso-Medium-v2-GGUF

@@ -1,173 +1,77 @@
 ---
-quantized_by: bartowski
-pipeline_tag: text-generation
-base_model: arcee-train/Virtuoso-Medium-v2
 tags:
 - mergekit
 - merge
 ---
-## Llamacpp imatrix Quantizations of Virtuoso-Medium-v2
-Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b4552">b4552</a> for quantization.
-Original model: https://huggingface.co/arcee-train/Virtuoso-Medium-v2
-All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8)
-Run them in [LM Studio](https://lmstudio.ai/)
-Run them directly with [llama.cpp](https://github.com/ggerganov/llama.cpp), or any other llama.cpp based project
-## Prompt format
-```
-<|im_start|>system
-{system_prompt}<|im_end|>
-<|im_start|>user
-{prompt}<|im_end|>
-<|im_start|>assistant
-```
-## Download a file (not the whole branch) from below:
-| Filename | Quant type | File Size | Split | Description |
-| -------- | ---------- | --------- | ----- | ----------- |
-| [Virtuoso-Medium-v2-Q8_0.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q8_0.gguf) | Q8_0 | 34.82GB | false | Extremely high quality, generally unneeded but max available quant. |
-| [Virtuoso-Medium-v2-Q6_K_L.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q6_K_L.gguf) | Q6_K_L | 27.26GB | false | Uses Q8_0 for embed and output weights. Very high quality, near perfect, *recommended*. |
-| [Virtuoso-Medium-v2-Q6_K.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q6_K.gguf) | Q6_K | 26.88GB | false | Very high quality, near perfect, *recommended*. |
-| [Virtuoso-Medium-v2-Q5_K_L.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q5_K_L.gguf) | Q5_K_L | 23.74GB | false | Uses Q8_0 for embed and output weights. High quality, *recommended*. |
-| [Virtuoso-Medium-v2-Q5_K_M.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q5_K_M.gguf) | Q5_K_M | 23.26GB | false | High quality, *recommended*. |
-| [Virtuoso-Medium-v2-Q5_K_S.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q5_K_S.gguf) | Q5_K_S | 22.64GB | false | High quality, *recommended*. |
-| [Virtuoso-Medium-v2-Q4_1.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q4_1.gguf) | Q4_1 | 20.64GB | false | Legacy format, similar performance to Q4_K_S but with improved tokens/watt on Apple silicon. |
-| [Virtuoso-Medium-v2-Q4_K_L.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q4_K_L.gguf) | Q4_K_L | 20.42GB | false | Uses Q8_0 for embed and output weights. Good quality, *recommended*. |
-| [Virtuoso-Medium-v2-Q4_K_M.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q4_K_M.gguf) | Q4_K_M | 19.85GB | false | Good quality, default size for most use cases, *recommended*. |
-| [Virtuoso-Medium-v2-Q4_K_S.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q4_K_S.gguf) | Q4_K_S | 18.78GB | false | Slightly lower quality with more space savings, *recommended*. |
-| [Virtuoso-Medium-v2-Q4_0.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q4_0.gguf) | Q4_0 | 18.71GB | false | Legacy format, offers online repacking for ARM and AVX CPU inference. |
-| [Virtuoso-Medium-v2-IQ4_NL.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ4_NL.gguf) | IQ4_NL | 18.68GB | false | Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference. |
-| [Virtuoso-Medium-v2-Q3_K_XL.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q3_K_XL.gguf) | Q3_K_XL | 17.92GB | false | Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability. |
-| [Virtuoso-Medium-v2-IQ4_XS.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ4_XS.gguf) | IQ4_XS | 17.69GB | false | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
-| [Virtuoso-Medium-v2-Q3_K_L.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q3_K_L.gguf) | Q3_K_L | 17.24GB | false | Lower quality but usable, good for low RAM availability. |
-| [Virtuoso-Medium-v2-Q3_K_M.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q3_K_M.gguf) | Q3_K_M | 15.93GB | false | Low quality. |
-| [Virtuoso-Medium-v2-IQ3_M.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ3_M.gguf) | IQ3_M | 14.81GB | false | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
-| [Virtuoso-Medium-v2-Q3_K_S.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q3_K_S.gguf) | Q3_K_S | 14.39GB | false | Low quality, not recommended. |
-| [Virtuoso-Medium-v2-IQ3_XS.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ3_XS.gguf) | IQ3_XS | 13.70GB | false | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
-| [Virtuoso-Medium-v2-Q2_K_L.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q2_K_L.gguf) | Q2_K_L | 13.07GB | false | Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable. |
-| [Virtuoso-Medium-v2-Q2_K.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-Q2_K.gguf) | Q2_K | 12.31GB | false | Very low quality but surprisingly usable. |
-| [Virtuoso-Medium-v2-IQ2_M.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ2_M.gguf) | IQ2_M | 11.26GB | false | Relatively low quality, uses SOTA techniques to be surprisingly usable. |
-| [Virtuoso-Medium-v2-IQ2_S.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ2_S.gguf) | IQ2_S | 10.39GB | false | Low quality, uses SOTA techniques to be usable. |
-| [Virtuoso-Medium-v2-IQ2_XS.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ2_XS.gguf) | IQ2_XS | 9.96GB | false | Low quality, uses SOTA techniques to be usable. |
-| [Virtuoso-Medium-v2-IQ2_XXS.gguf](https://huggingface.co/bartowski/Virtuoso-Medium-v2-GGUF/blob/main/Virtuoso-Medium-v2-IQ2_XXS.gguf) | IQ2_XXS | 9.03GB | false | Very low quality, uses SOTA techniques to be usable. |
-## Embed/output weights
-Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.
-## Downloading using huggingface-cli
-<details>
-  <summary>Click to view download instructions</summary>
-First, make sure you have hugginface-cli installed:
 ```
-pip install -U "huggingface_hub[cli]"
-```
-Then, you can target the specific file you want:
-```
-huggingface-cli download bartowski/Virtuoso-Medium-v2-GGUF --include "Virtuoso-Medium-v2-Q4_K_M.gguf" --local-dir ./
-```
-If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
-```
-huggingface-cli download bartowski/Virtuoso-Medium-v2-GGUF --include "Virtuoso-Medium-v2-Q8_0/*" --local-dir ./
-```
-You can either specify a new local-dir (Virtuoso-Medium-v2-Q8_0) or download them all in place (./)
-</details>
-## ARM/AVX information
-Previously, you would download Q4_0_4_4/4_8/8_8, and these would have their weights interleaved in memory in order to improve performance on ARM and AVX machines by loading up more data in one pass.
-Now, however, there is something called "online repacking" for weights. details in [this PR](https://github.com/ggerganov/llama.cpp/pull/9921). If you use Q4_0 and your hardware would benefit from repacking weights, it will do it automatically on the fly.
-As of llama.cpp build [b4282](https://github.com/ggerganov/llama.cpp/releases/tag/b4282) you will not be able to run the Q4_0_X_X files and will instead need to use Q4_0.
-Additionally, if you want to get slightly better quality for , you can use IQ4_NL thanks to [this PR](https://github.com/ggerganov/llama.cpp/pull/10541) which will also repack the weights for ARM, though only the 4_4 for now. The loading time may be slower but it will result in an overall speed incrase.
-<details>
-  <summary>Click to view Q4_0_X_X information (deprecated</summary>
-I'm keeping this section to show the potential theoretical uplift in performance from using the Q4_0 with online repacking.
-<details>
-  <summary>Click to view benchmarks on an AVX2 system (EPYC7702)</summary>
-| model                          |       size |     params | backend    | threads |          test |                  t/s |  % (vs Q4_0)  |
-| ------------------------------ | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |-------------: |
-| qwen2 3B Q4_0                  |   1.70 GiB |     3.09 B | CPU        |      64 |         pp512 |        204.03 ± 1.03 |          100% |
-| qwen2 3B Q4_0                  |   1.70 GiB |     3.09 B | CPU        |      64 |        pp1024 |        282.92 ± 0.19 |          100% |
-| qwen2 3B Q4_0                  |   1.70 GiB |     3.09 B | CPU        |      64 |        pp2048 |        259.49 ± 0.44 |          100% |
-| qwen2 3B Q4_0                  |   1.70 GiB |     3.09 B | CPU        |      64 |         tg128 |         39.12 ± 0.27 |          100% |
-| qwen2 3B Q4_0                  |   1.70 GiB |     3.09 B | CPU        |      64 |         tg256 |         39.31 ± 0.69 |          100% |
-| qwen2 3B Q4_0                  |   1.70 GiB |     3.09 B | CPU        |      64 |         tg512 |         40.52 ± 0.03 |          100% |
-| qwen2 3B Q4_K_M                |   1.79 GiB |     3.09 B | CPU        |      64 |         pp512 |        301.02 ± 1.74 |          147% |
-| qwen2 3B Q4_K_M                |   1.79 GiB |     3.09 B | CPU        |      64 |        pp1024 |        287.23 ± 0.20 |          101% |
-| qwen2 3B Q4_K_M                |   1.79 GiB |     3.09 B | CPU        |      64 |        pp2048 |        262.77 ± 1.81 |          101% |
-| qwen2 3B Q4_K_M                |   1.79 GiB |     3.09 B | CPU        |      64 |         tg128 |         18.80 ± 0.99 |           48% |
-| qwen2 3B Q4_K_M                |   1.79 GiB |     3.09 B | CPU        |      64 |         tg256 |         24.46 ± 3.04 |           83% |
-| qwen2 3B Q4_K_M                |   1.79 GiB |     3.09 B | CPU        |      64 |         tg512 |         36.32 ± 3.59 |           90% |
-| qwen2 3B Q4_0_8_8              |   1.69 GiB |     3.09 B | CPU        |      64 |         pp512 |        271.71 ± 3.53 |          133% |
-| qwen2 3B Q4_0_8_8              |   1.69 GiB |     3.09 B | CPU        |      64 |        pp1024 |       279.86 ± 45.63 |          100% |
-| qwen2 3B Q4_0_8_8              |   1.69 GiB |     3.09 B | CPU        |      64 |        pp2048 |        320.77 ± 5.00 |          124% |
-| qwen2 3B Q4_0_8_8              |   1.69 GiB |     3.09 B | CPU        |      64 |         tg128 |         43.51 ± 0.05 |          111% |
-| qwen2 3B Q4_0_8_8              |   1.69 GiB |     3.09 B | CPU        |      64 |         tg256 |         43.35 ± 0.09 |          110% |
-| qwen2 3B Q4_0_8_8              |   1.69 GiB |     3.09 B | CPU        |      64 |         tg512 |         42.60 ± 0.31 |          105% |
-Q4_0_8_8 offers a nice bump to prompt processing and a small bump to text generation
-</details>
-</details>
-## Which file should I choose?
-<details>
-  <summary>Click here for details</summary>
-A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
-The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
-If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
-If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
-Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
-If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
-If you want to get more into the weeds, you can check out this extremely useful feature chart:
-[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
-But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
-These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
-The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
-</details>
-## Credits
-Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset.
-Thank you ZeroWw for the inspiration to experiment with embed/output.
-Thank you to LM Studio for sponsoring my work.

 ---
+base_model:
+- Qwen/Qwen2.5-32B
+library_name: transformers
 tags:
 - mergekit
 - merge
 ---
+<img src="https://huggingface.co/arcee-train/Virtuoso-Medium-v2/resolve/main/virtuoso-medium.jpg" alt="Virtuoso-Lite Logo" style="display: block; margin: 0 auto;" />
+**Virtuoso-Medium-v2 (32B)** is our next-generation, 32-billion-parameter language model that builds upon the original Virtuoso-Medium architecture. This version is distilled from Deepseek-v3, leveraging an expanded dataset of 5B+ tokens worth of logits. It achieves higher benchmark scores than our previous release (including surpassing Arcee-Nova 2024 in certain tasks).
+### Model Details
+- **Architecture Base:** Qwen-2.5-32B
+- **Parameter Count:** 32B
+- **Tokenizer:**
+  - Initially integrated with Deepseek-v3 tokenizer for logit extraction.
+  - Final alignment uses the Qwen tokenizer, using specialized “tokenizer surgery” for cross-architecture compatibility.
+- **Distillation Data:**
+  - ~1.1B tokens/logits from Deepseek-v3’s training data.
+  - Logit-level distillation using a proprietary “fusion merging” approach afterwards for maximum fidelity.
+- **License:** [Apache-2.0](#license)
+### Background on Deepseek Distillation
+Deepseek-v3 serves as the teacher model, from which we capture logits across billions of tokens. Rather than standard supervised fine-tuning, we apply a full logit-level replication. This ensures more precise transference of knowledge, including advanced reasoning in:
+- Technical and scientific queries
+- Complex code generation
+- Mathematical problem-solving
+### Intended Use Cases
+- **Advanced Chatbots & Virtual Assistants**
+- **Enterprise Data Analysis & Workflow Automation**
+- **Research Simulations & Natural Language Understanding**
+- **Educational Tools for STEM Fields**
+### Evaluations
+<img src="https://huggingface.co/arcee-train/Virtuoso-Lite/resolve/main/Benchmarks.png" alt="Virtuoso-Lite Logo" style="display: block; margin: 0 auto;" />
+### How to Use
+Below is a sample code snippet using `transformers`:
+```python
+from transformers import AutoTokenizer, AutoModelForCausalLM
+model_name = "arcee-ai/virtuoso-medium-v2-32b"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name)
+prompt = "Provide a concise summary of quantum entanglement."
+inputs = tokenizer(prompt, return_tensors="pt")
+outputs = model.generate(**inputs, max_new_tokens=150)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
+### Training & Fine-Tuning
+- **Initial Training:** Began with Qwen-32B, calibrated for large-scale text ingestion.
+- **Distillation & Merging:**
+  - Trained on ~1.1B tokens worth of Deepseek-v3 logits.
+  - Employed “fusion merging” to retain as much teacher expertise as possible.
+  - Final step included DPO to improve alignment and reduce model hallucinations.
+- **Continuous Development:** Additional R1 distillations are in progress to further enhance performance and specialization.
+### Performance
+Thanks to a larger parameter count and a richer training corpus, Virtuoso-Medium-v2 delivers high scores across multiple benchmarks (BBH, MMLU-PRO, MATH, etc.). It frequently surpasses other 30B+ models and even some 70B+ architectures in specific tasks.
+### Limitations
+- **Context Length:** 128k Tokens
+- **Knowledge Cut-off:** Training data may not reflect the latest events or developments, leading to gaps in current knowledge beyond June 2024.
+### Ethical Considerations
+- **Content Generation Risks:** Like any language model, Virtuoso-Medium-v2 can potentially generate harmful or biased content if prompted in certain ways.
+### License
+**Virtuoso-Medium-v2 (32B)** is released under the [Apache-2.0 License](https://www.apache.org/licenses/LICENSE-2.0). You are free to use, modify, and distribute this model in both commercial and non-commercial applications, subject to the terms and conditions of the license.
+If you have questions or would like to share your experiences using these models, please connect with us on social media. We’re excited to see what you build—and how these models help you innovate!