Update README.md

Add GGUF internal file structure
2025-07-01 09:07:31 +01:00 · 2025-07-01 08:59:38 +01:00 · 2025-07-01 08:19:46 +01:00 · 2025-07-01 08:18:25 +01:00 · 2025-07-01 07:38:42 +01:00 · 2025-07-01 07:31:37 +01:00
97 changed files with 17049 additions and 2 deletions
--- a/Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf
+++ b/Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0aef3c959bb993ff6b3436ceba8f6c84380632e35024212463d2c1b9d87c649a
+size 9551090944
--- a/Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf
+++ b/Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:225dbd86c81f9f0c21e983dc48ae50a4030438870b06775036af68b413d9f491
+size 9258390784
--- a/Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf
+++ b/Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:065931827062c69435d6070cb52d0756d825f44dedada29d3e37393ffeb219b6
+size 11590161664
--- a/Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf
+++ b/Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5dc5bfdfe5688dae240596d4c7b5ce27596f6e8b04cc13acc93a024592ef6acc
+size 10801190144
--- a/Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf
+++ b/Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89519ef728550abb6337361e574ada6bff595979a2bcd25f47d4a45c161778e5
+size 9932674304
--- a/Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf
+++ b/Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd7f9a8b813884de5d97574e72696f09db90bfdc61c06b33caf196bfc10a9bc6
+size 8857474304
--- a/README.md
+++ b/README.md
@@ -17,6 +17,211 @@ tags:

 # Experimental layer-wise + pruned (layers 4 and 5) quantization of cognitivecomputations/Dolphin-Mistral-24B-Venice-Edition

-## Upload in progress...
+Using [LLaMA C++][llm] release [b5770][llm-rel] for quantization.

-Full model card will be available once all files have been uploaded
+Original model: [cognitivecomputations/Dolphin-Mistral-24B-Venice-Edition][mdl]
+
+From the original model creators:
+
+> [![Discord](https://img.shields.io/discord/1156064224225808488?logo=Discord&logoColor=%23ffffff&label=Discord&link=https%3A%2F%2Fdiscord.gg%2FtCMkMDDHwm)](https://discord.gg/h3K4XGj2RH)
+> Discord: https://discord.gg/h3K4XGj2RH  
+> Website: https://dphn.ai  
+> Twitter: https://x.com/dphnAI  
+>
+> ![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/68485b28c949339ca04c370c/LMOLMYwK-ixnGGdSBXew6.jpeg)
+>
+> ## What is Dolphin Mistral 24B Venice Edition?
+>
+> Dolphin Mistral 24B Venice Edition is a collaborative project we undertook with Venice.ai with the goal of creating the most uncensored version of Mistral 24B for use within the Venice ecosystem.
+>
+> Dolphin Mistral 24B Venice Edition is now live on https://venice.ai/ as “Venice Uncensored,” the new default model for all Venice users.
+>
+> Dolphin aims to be a general purpose model, similar to the models behind ChatGPT, Claude, Gemini. But these models present problems for businesses seeking to include AI in their products.
+> 1) They maintain control of the system prompt, deprecating and changing things as they wish, often causing software to break.
+> 2) They maintain control of the model versions, sometimes changing things silently, or deprecating older models that your business relies on.
+> 3) They maintain control of the alignment, and in particular the alignment is one-size-fits all, not tailored to the application.
+> 4) They can see all your queries and they can potentially use that data in ways you wouldn't want.
+> Dolphin, in contrast, is steerable and gives control to the system owner. You set the system prompt.  You decide the alignment.  You have control of your data.  Dolphin does not impose its ethics or guidelines on you.  You are the one who decides the guidelines.
+>
+> Dolphin belongs to YOU, it is your tool, an extension of your will.
+> Just as you are personally responsible for what you do with a knife, gun, fire, car, or the internet, you are the creator and originator of any content you generate with Dolphin.
+
+From Eric Hartford's, the creator of the Dolphin model series, [Uncensored Models](https://erichartford.com/uncensored-models):
+
+> Most of these models (for example, Alpaca, Vicuna, WizardLM, MPT-7B-Chat, Wizard-Vicuna, GPT4-X-Vicuna) have some sort of embedded alignment. For general purposes, this is a good thing. This is what stops the model from doing bad things, like teaching you how to cook meth and make bombs. But what is the nature of this alignment? And, why is it so?
+>
+> The reason these models are aligned is that they are trained with data that was generated by ChatGPT, which itself is aligned by an alignment team at OpenAI. As it is a black box, we don't know all the reasons for the decisions that were made, but we can observe it generally is aligned with American popular culture, and to obey American law...
+
+# PLEASE READ THIS BEFORE USING THESE EXPERIMENTAL VERSIONS!
+
+An area of personal interest is finding ways to optimize the inference performance of LLMs when deployed in resource-constrained environments like commodity hardware, desktops, laptops, mobiles, edge devices, etc. There are many approaches to accomplish this, including architecture simplification and knowledge distillation, but my focus has been primarily on quantization and pruning.
+
+The method used to produce these experimental versions is covered in [Squeezing Tensor Bits: the quest for smaller LLMs][mdm], but at a high level it involves using a custom version of `llama-imatrix` and `llama-quantize` to identify influential tensors, quantize the most important layers to higher bit precision and the less important to lower bits, and remove (prune) one or more layers. This process was partly inspired by Dumitru's et al [Layer-Wise Quantization: A Pragmatic and Effective Method for Quantizing LLMs Beyond Integer Bit-Levels][lwq-ppr], and Xin Men's et al [ShortGPT: Layers in Large Language Models are More Redundant Than You Expect][sgpt-ppr]
+
+As of version [b5125][qtz-lwq], [llama-quantize][qtz] can perform **tensor-wide quantization (TWQ)**, whereby user-defined tensors are quantized at a specific level, or perform **layer-wise quantization (LWQ)** by selecting different quantization types per tensor/layer. For example, `--tensor-type attn_v=q6_k` will quantize all *Attention Value* tensors at *q6_k* (TWQ), and `--tensor-type "\.([0-9]|1[01257]|31)\.attn_k=q4_k"` will quantize *Attention Key* tensors on layers 0 to 9, 10, 11, 12, 15, 17 and 31 at *q4_k*, leaving the remaining layers at their default value (LWQ).
+
+As of version [b5740][qtz-prn], [llama-quantize][qtz] can also prune models during quantisation by providing a comma-separated list in the `--prune-layers` command line option. The pruning operation will renumber remaining layers to avoid gaps in the sequence, update the relevant model metadata and, if an imatrix is available, it will use the correct importance score vector. This option can be used alongside `--tensor-type` to perform tensor/layer-wise quantization on selected tensor types, whilst at the same time pruning others. For example:
+```
+llama-quantize --tensor-type attn=q6_k --prune-layers 3,7,11 --imatrix imatrix.dat model-f32.gguf model-q4_k_m.gguf q4_k_m
+```
+
+An enhanced version of [llama-imatrix][imx] generates useful statistics to guide the tensor and layer selection process. `--show-statistics` will display:
+
+- **Σ(Act²):** the sum of all squared activations over the tensor (i.e. the Importance Scores)
+- **Min & Max:** minimum and maximum squared activation values
+- **μ & σ:** activations' mean and standard deviation
+- **% Active:** proportion of elements whose average squared activation exceeds a very small threshold (1e-5). Helpful to determine how alive/dormant the tensor is during inference
+- **N:** number of squared activations in the tensor
+- **Entropy:** entropy of the squared activation distribution, in bits (standard Shannon entropy measurement)
+- **E (norm):** Normalized entropy.
+- **ZD Score:** z-score distribution as described in 3.1 Layer Importance Scores in the Layer-Wise Quantization paper
+- **CosSim:** cosine similarity between same type tensors with respect to the previous layer (i.e. blk.7.attn_k and blk.6.attn_k)
+
+Please note that statistics are calculated for each individual tensor and should be used to compare between tensors of the same type only. For example, assuming that *attn_k* in layer 10 has a higher influence during inference than *attn_k* in layer 7 because its **Σ(Act²)** is larger makes sense, whilst concluding the same between *attn_k* and *ffn_down* does not.
+
+There’s a [pull request][imtx-pr] to merge these changes back into the core llama.cpp project. This may or may not ever happen so, until then, the modified version will be available on [GitHub][gh].
+
+For testing and comparison I use models produced by [Unsloth][ust] ([Daniel and Michael Han][ust-ai] do some really advanced level stuff!) and [Bartowski][btk] (see credits below) but if they don't provide versions of the required model, all tests and comparisons are done against naive quantizations obtained by simply running `llama-quantize` with no further optimization.
+
+All experimental versions were generated using an appropriate imatrix created from calibration datasets available at [eaddario/imatrix-calibration][ical]. At its core, an Importance Matrix (imatrix) is a table or, more broadly, a structured representation that scores the relative importance of different features or parameters in a machine learning model. It essentially quantifies the "impact" each feature has on a specific outcome, prediction, or relationship being modelled, and it helps to counterbalance the negative effects of quantization and pruning.
+
+The process to generate these models is roughly as follows:
+
+1. Convert the original model's tensors to [GGUF][ggf] F16*
+2. Estimate the Perplexity score for the F16 model (baseline) using the [wikitext-2-raw-v1][wki-dat] dataset, and save the [logits][lgt]
+3. Generate an [imatrix][imx-dat] from selected calibration datasets
+4. Determine tensor and layer Importance Score contribution using the enhanced version of `llama-imatrix`
+5. Select an appropriate quant level for each tensor and quantize/prune the model using `llama-quantize`. In this model's case, layers **4** and **5** have been pruned
+6. Calculate Perplexity, KL Divergence, ARC (Easy+Challenge), HellaSwag, MMLU, Truthful QA and WinoGrande scores for each quantized model
+7. Keep versions with the best scores
+8. Repeat until all desired quants are created. I find that quantizations below Q3/IQ3 are not fit for my purposes and therefore do not usually generate them, but happy to provide other quants on request.
+
+*[BF16][bf16] would be preferred, but Apple's GPUs don't support it yet, and therefore any operations are executed in the CPU, making it unacceptably slow. This is expected to change in the near term but until then, if you are using Apple kit avoid using any models tagged BF16
+
+# Models
+
+### Sizes (in GB)
+| Model                                                                                                       | Bartowski | Repo | Shrinkage |
+| ----------------------------------------------------------------------------------------------------------- | --------: | ---: | --------: |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf)   |      10.7 |  9.6 |     10.3% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf)   |       9.9 |  9.3 |      6.2% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf) |      13.5 | 11.6 |     14.1% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf) |      12.4 | 10.8 |     12.9% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf) |      11.5 |  9.9 |     13.9% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf) |      10.4 |  8.9 |     14.4% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf) |      14.3 | 12.4 |     13.3% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf) |      13.5 | 11.7 |     13.3% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf) |      16.8 | 14.3 |     14.9% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf) |      16.3 | 13.9 |     14.7% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf)     |      19.7 | 16.8 |     14.7% |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf)     |      25.1 | 21.9 |     12.7% |
+
+### Perplexity and KL Divergence scores
+| Model                                                                                                       |                μPPL |   𝜌PPL |               μKLD |        RMS Δp |
+| ----------------------------------------------------------------------------------------------------------- | ------------------: | -----: | -----------------: | ------------: |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf)   | 20.379006 ±0.160275 | 73.93% | 1.290608 ±0.004304 | 37.928 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf)   | 21.165413 ±0.164512 | 73.80% | 1.340446 ±0.004301 | 38.586 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf) | 18.783744 ±0.146959 | 74.79% | 1.199318 ±0.004258 | 36.745 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf) | 19.313300 ±0.150799 | 74.61% | 1.248712 ±0.004216 | 37.260 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf) | 18.723777 ±0.145380 | 75.90% | 1.226150 ±0.004006 | 36.807 ±0.087 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf) | 19.765437 ±0.153182 | 74.13% | 1.295119 ±0.004177 | 38.004 ±0.087 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf) | 18.556910 ±0.145472 | 74.92% | 1.187728 ±0.004237 | 36.521 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M-bartowski][b-q4km]                                        |  6.304728 ±0.042418 | 99.60% | 0.016941 ±0.000138 |  4.031 ±0.037 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf) | 18.663517 ±0.146425 | 74.87% | 1.192878 ±0.004250 | 36.598 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf) | 18.174846 ±0.142320 | 75.14% | 1.159685 ±0.004238 | 36.214 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf) | 18.199918 ±0.142513 | 75.20% | 1.160040 ±0.004229 | 36.220 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf)     | 18.213825 ±0.142965 | 75.05% | 1.158026 ±0.004262 | 36.219 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf)     | 18.203515 ±0.142826 | 75.02% | 1.158351 ±0.004265 | 36.227 ±0.088 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-F16](./Dolphin-Mistral-24B-Venice-Edition-F16.gguf)       |  6.180577 ±0.041038 |   100% |                N/A |           N/A |
+
+### ARC, HellaSwag, MMLU, Truthful QA and WinoGrande scores
+Scores generated using [llama-perplexity][ppl] with 750 tasks per test, and a context size of 768 tokens.
+
+For the test data used in the generation of these scores, follow the appropriate links: [HellaSwag][hsw-tst], [ARC, MMLU, Truthful QA][tst-dat] and [WinoGrande][wng-tst]
+
+| Model                                                                                                       |             ARC | HellaSwag |            MMLU |     Truthful QA |      WinoGrande | Avg Score |
+| ----------------------------------------------------------------------------------------------------------- | --------------: | --------: | --------------: | --------------: | --------------: | --------: |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf)   | 65.6000 ±1.7358 |     79.60 | 42.9333 ±1.8086 | 38.4000 ±1.7771 | 72.4000 ±1.6334 |     59.79 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf)   | 64.9333 ±1.7436 |     79.87 | 42.0000 ±1.8034 | 38.0000 ±1.7736 | 72.5333 ±1.6309 |     59.47 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL](./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf) | 68.4000 ±1.6988 |     80.66 | 44.9333 ±1.8176 | 38.1333 ±1.7748 | 74.4000 ±1.5947 |     61.31 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf) | 67.2000 ±1.7155 |     80.27 | 43.2000 ±1.8100 | 39.6000 ±1.7870 | 72.9333 ±1.6235 |     60.64 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf) | 66.6667 ±1.7225 |     80.67 | 43.8667 ±1.8132 | 39.4667 ±1.7860 | 72.2667 ±1.6358 |     60.59 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf) | 66.2667 ±1.7276 |     78.93 | 43.7333 ±1.8126 | 38.1333 ±1.7748 | 72.8000 ±1.6260 |     59.97 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf) | 68.0000 ±1.7045 |     80.93 | 45.2000 ±1.8185 | 36.6667 ±1.7608 | 72.1333 ±1.6382 |     60.59 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M-bartowski][b-q4km]                                        | 69.8667 ±1.6766 |     84.27 | 45.3333 ±1.8190 | 37.6000 ±1.7699 | 80.2667 ±1.4542 |     63.47 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf) | 67.0667 ±1.7172 |     81.07 | 45.2000 ±1.8185 | 36.2667 ±1.7567 | 72.0000 ±1.6406 |     60.32 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf) | 67.0667 ±1.7172 |     81.73 | 44.5333 ±1.8160 | 37.8667 ±1.7724 | 73.8667 ±1.6054 |     61.01 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf) | 67.3333 ±1.7137 |     81.47 | 44.2667 ±1.8149 | 38.6667 ±1.7794 | 74.2667 ±1.5974 |     61.20 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf)     | 67.4667 ±1.7119 |     81.07 | 44.5333 ±1.8160 | 39.6000 ±1.7870 | 73.8667 ±1.6054 |     61.31 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf)     | 68.1333 ±1.7026 |     81.33 | 44.9333 ±1.8176 | 38.2667 ±1.7759 | 74.4000 ±1.5947 |     61.41 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-F16](./Dolphin-Mistral-24B-Venice-Edition-F16.gguf)       | 70.8000 ±1.6614 |     84.53 | 45.3333 ±1.8190 | 38.1333 ±1.7748 | 80.2667 ±1.4542 |     63.81 |
+
+### Tokens per Second - Benchmarks
+Scores generated using [llama-bench][bch]. Naive (`llama-quantize` with no optimization) Q4_K_M quantization included for comparison.
+
+| model                                                                                                       |      size |  params | backend    | threads |          test |           t/s |
+| ----------------------------------------------------------------------------------------------------------- | --------: | ------: | ---------- | ------: | ------------: | ------------: |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf) | 11.53 GiB | 22.46 B | Metal,BLAS |      12 |         pp512 | 266.57 ±14.60 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf) | 11.53 GiB | 22.46 B | Metal,BLAS |      12 |         tg128 |   27.60 ±0.54 |
+| [Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M](./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf) | 11.53 GiB | 22.46 B | Metal,BLAS |      12 | pp1024+tg1024 |   41.55 ±2.99 |
+| [Dolphin-Mistral-24B-Venice-Edition-Q4_K_M-bartowski][b-q4km]                                               | 13.34 GiB | 23.57 B | Metal,BLAS |      12 |         pp512 |  253.67 ±2.26 |
+| [Dolphin-Mistral-24B-Venice-Edition-Q4_K_M-bartowski][b-q4km]                                               | 13.34 GiB | 23.57 B | Metal,BLAS |      12 |         tg128 |   27.69 ±0.46 |
+| [Dolphin-Mistral-24B-Venice-Edition-Q4_K_M-bartowski][b-q4km]                                               | 13.34 GiB | 23.57 B | Metal,BLAS |      12 | pp1024+tg1024 |   45.54 ±0.18 |
+
+# Metrics used
+**[Perplexity][ppx]:** one of the key metrics used in NLP evaluation. It measures the quality of a language model by evaluating how well it predicts the next token given a particular sequence of words. A PPL of **1** indicates an exact match between predicted and actual, whereas values greater than one indicate a degree of "surprise" the generated token differs from the expected.
+
+**[Kullback–Leibler (KL) Divergence][kld]:** a statistical measure of how much a probability distribution differs from another. When quantizing models (or altering the original tensors in any way for that matter), the closest we can preserve the weights' probability distribution to the original model the better, thus the closest to **0** the better.
+
+**[AI2 Reasoning Challenge (ARC)][arc]:** a benchmark to evaluate the ability of AI models to answer complex science questions that require logical reasoning beyond pattern matching.
+
+**[HellaSwag][hsw]:** the Harder Endings, Longer contexts, and Low-shot Activities for Situations With Adversarial Generations (bit of a mouthful!) is a benchmark designed to test commonsense natural language inference. It requires the model to predict the most likely ending of a sentence.
+
+**[MMLU][mmlu]:** the Massive Multitask Language Understanding evaluates LLMs’ general knowledge and problem-solving abilities across 57 subjects, including elementary mathematics, US history, computer science, and law.
+
+**[Truthful QA][tqa]:** evaluates how well LLMs generate truthful responses to questions. It identifies whether AI models can avoid generating false or misleading information, particularly in areas where human knowledge is prone to misconceptions.
+
+**[Winogrande][wng]:** based on the [Winograd Schema Challenge][wng-chl], is a natural language understanding task requiring models to resolve ambiguities in sentences involving pronoun references.
+
+## Credits
+[LLaMa C++][llm] has a large and vibrant community of [contributors][llm-ctt] (~1,200 last time I checked) that actively maintain and extend its functionality, adding new models and architectures almost as fast as they appear (considering the breakneck speed at which the AI/ML field is advancing, this alone is a remarkable feat!), and whilst I'm grateful to each and everyone of them, I want to recognise three people in particular: **Thank You!** [Colin Kealty][btk] for the many contributions and for being one of the best sources of high quality quantized models available on Hugging Face, and a really big ***Thank You!*** to [Georgi Gerganov][ggg] for his amazing work with **llama.cpp** and the **ggml/gguf** libraries, and [Iwan Kawrakow][ikk] for being one of the key authors behind the many quantisation algorithms and the imatrix functionality.
+
+[arc]:      https://leaderboard.allenai.org/arc/submissions/get-started
+[b-q4km]:   https://huggingface.co/bartowski/cognitivecomputations_Dolphin-Mistral-24B-Venice-Edition-GGUF/blob/main/cognitivecomputations_Dolphin-Mistral-24B-Venice-Edition-Q4_K_M.gguf
+[bch]:      https://github.com/ggml-org/llama.cpp/tree/master/tools/llama-bench
+[bf16]:     https://en.wikipedia.org/wiki/Bfloat16_floating-point_format
+[btk]:      https://huggingface.co/bartowski
+[ggf]:      https://huggingface.co/docs/hub/en/gguf
+[ggg]:      https://github.com/ggerganov
+[gh]:       https://github.com/EAddario/llama.cpp/tree/imatrix
+[hsw-tst]:  https://github.com/klosax/hellaswag_text_data
+[hsw]:      https://rowanzellers.com/hellaswag
+[ical]:     https://huggingface.co/datasets/eaddario/imatrix-calibration
+[ikk]:      https://github.com/ikawrakow
+[imtx-pr]:  https://github.com/ggml-org/llama.cpp/pull/12718
+[imx-dat]:  https://huggingface.co/eaddario/Dolphin-Mistral-24B-Venice-Edition-pruned-GGUF/tree/main/imatrix
+[imx]:      https://github.com/ggml-org/llama.cpp/tree/master/tools/imatrix
+[kld]:      https://en.wikipedia.org/wiki/Kullback–Leibler_divergence
+[lgt]:      https://huggingface.co/eaddario/Dolphin-Mistral-24B-Venice-Edition-pruned-GGUF/tree/main/logits
+[llm-ctt]:  https://github.com/ggml-org/llama.cpp/graphs/contributors
+[llm-rel]:  https://github.com/ggml-org/llama.cpp/releases/tag/b5770
+[llm]:      https://github.com/ggerganov/llama.cpp
+[lwq-ppr]:  https://arxiv.org/abs/2406.17415
+[mdl]:      https://huggingface.co/cognitivecomputations/Dolphin-Mistral-24B-Venice-Edition
+[mdm]:      https://medium.com/@eaddario/squeezing-tensor-bits-the-quest-for-smaller-llms-86b23bd052ca
+[mmlu]:     https://github.com/hendrycks/test
+[ppl]:      https://github.com/ggml-org/llama.cpp/tree/master/tools/perplexity
+[ppx]:      https://huggingface.co/docs/transformers/en/perplexity
+[qtz-lwq]:  https://github.com/ggml-org/llama.cpp/releases/tag/b5125
+[qtz-prn]:  https://github.com/ggml-org/llama.cpp/releases/tag/b5740
+[qtz]:      https://github.com/ggml-org/llama.cpp/tree/master/tools/quantize
+[sgpt-ppr]: https://arxiv.org/abs/2403.03853
+[tqa]:      https://github.com/sylinrl/TruthfulQA
+[tst-dat]:  https://huggingface.co/datasets/ikawrakow/validation-datasets-for-llama.cpp/tree/main
+[u-q4km]:   https://huggingface.co/unsloth/
+[ust-ai]:   https://unsloth.ai
+[ust]:      https://huggingface.co/unsloth
+[wki-dat]:  https://huggingface.co/datasets/Salesforce/wikitext/tree/main/wikitext-2-raw-v1
+[wng-chl]:  https://cdn.aaai.org/ocs/4492/4492-21843-1-PB.pdf
+[wng-tst]:  https://huggingface.co/datasets/ikawrakow/winogrande-eval-for-llama.cpp/tree/main
+[wng]:      https://winogrande.allenai.org
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-F16.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-F16.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 39 key-value pairs and 363 tensors from ./Dolphin-Mistral-24B-Venice-Edition-F16.gguf (version GGUF V3 (latest))
+
+Final result: 70.8000 +/- 1.6614
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =   40685.08 ms
+llama_perf_context_print: prompt eval time =  151474.18 ms / 36666 tokens (    4.13 ms per token,   242.06 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  152313.64 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-F16.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-F16.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 39 key-value pairs and 363 tensors from ./Dolphin-Mistral-24B-Venice-Edition-F16.gguf (version GGUF V3 (latest))
+
+750	84.53333333%	[81.7702%, 86.9445%]
+
+
+llama_perf_context_print:        load time =    2898.86 ms
+llama_perf_context_print: prompt eval time =  545800.60 ms / 129319 tokens (    4.22 ms per token,   236.93 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  549439.05 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-F16.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-F16.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-F16.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-F16.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 39 key-value pairs and 363 tensors from ./Dolphin-Mistral-24B-Venice-Edition-F16.gguf (version GGUF V3 (latest))
+
+Final result: 45.3333 +/- 1.8190
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =    2912.06 ms
+llama_perf_context_print: prompt eval time =  289195.06 ms / 68956 tokens (    4.19 ms per token,   238.44 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  290438.95 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-F16.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-F16.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 39 key-value pairs and 363 tensors from ./Dolphin-Mistral-24B-Venice-Edition-F16.gguf (version GGUF V3 (latest))
+
+Final result: 38.1333 +/- 1.7748
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =    2866.02 ms
+llama_perf_context_print: prompt eval time =  221192.17 ms / 51053 tokens (    4.33 ms per token,   230.81 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  222713.65 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-F16.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-F16.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 39 key-value pairs and 363 tensors from ./Dolphin-Mistral-24B-Venice-Edition-F16.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 80.2667 +/- 1.4542
+
+llama_perf_context_print:        load time =    2915.27 ms
+llama_perf_context_print: prompt eval time =   93765.90 ms / 22541 tokens (    4.16 ms per token,   240.40 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =   94258.84 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-IQ3_M.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-IQ3_M.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-IQ3_S.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-IQ3_S.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-IQ4_NL.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-IQ4_NL.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q3_K_L.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q3_K_L.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q3_K_M.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q3_K_M.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q3_K_S.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q3_K_S.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q4_K_M.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q4_K_M.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q4_K_S.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q4_K_S.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q5_K_M.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q5_K_M.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q5_K_S.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q5_K_S.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q6_K.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q6_K.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-Q8_0.md
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-Q8_0.md
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf (version GGUF V3 (latest))
+
+Final result: 65.6000 +/- 1.7358
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    4337.33 ms
+llama_perf_context_print: prompt eval time =  154185.52 ms / 36666 tokens (    4.21 ms per token,   237.80 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  155064.54 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf (version GGUF V3 (latest))
+
+750	79.60000000%	[76.5686%, 82.3297%]
+
+
+llama_perf_context_print:        load time =     748.01 ms
+llama_perf_context_print: prompt eval time =  545228.12 ms / 129319 tokens (    4.22 ms per token,   237.18 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  549120.81 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf (version GGUF V3 (latest))
+
+Final result: 42.9333 +/- 1.8086
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     733.03 ms
+llama_perf_context_print: prompt eval time =  286965.35 ms / 68956 tokens (    4.16 ms per token,   240.29 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  288291.52 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  20.379006 ±   0.160275
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  73.93%
+Mean ln(PPL(Q)/PPL(base))     :   1.193094 ±   0.005360
+Mean PPL(Q)/PPL(base)         :   3.297266 ±   0.017673
+Mean PPL(Q)-PPL(base)         :  14.198428 ±   0.132841
+
+====== KL divergence statistics ======
+Mean    KLD:   1.290608 ±   0.004304
+Maximum KLD:  27.217335
+99.9%   KLD:  13.970652
+99.0%   KLD:   8.700209
+99.0%   KLD:   8.700209
+Median  KLD:   0.800781
+10.0%   KLD:   0.078073
+ 5.0%   KLD:   0.028142
+ 1.0%   KLD:   0.004895
+Minimum KLD:   0.000072
+
+====== Token probability statistics ======
+Mean    Δp: -18.226 ± 0.085 %
+Maximum Δp: 95.351%
+99.9%   Δp: 73.378%
+99.0%   Δp: 48.957%
+95.0%   Δp: 22.742%
+90.0%   Δp:  9.547%
+75.0%   Δp:  0.003%
+Median  Δp: -5.000%
+25.0%   Δp: -33.511%
+10.0%   Δp: -75.041%
+ 5.0%   Δp: -91.958%
+ 1.0%   Δp: -99.782%
+ 0.1%   Δp: -99.968%
+Minimum Δp: -99.998%
+RMS Δp    : 37.928 ± 0.088 %
+Same top p: 62.059 ± 0.125 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf (version GGUF V3 (latest))
+
+Final result: 38.4000 +/- 1.7771
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     775.72 ms
+llama_perf_context_print: prompt eval time =  219416.23 ms / 51053 tokens (    4.30 ms per token,   232.68 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  221003.17 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_m.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_M.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 72.4000 +/- 1.6334
+
+llama_perf_context_print:        load time =     754.95 ms
+llama_perf_context_print: prompt eval time =   95407.56 ms / 22541 tokens (    4.23 ms per token,   236.26 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =   95958.45 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf (version GGUF V3 (latest))
+
+Final result: 64.9333 +/- 1.7436
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    4191.63 ms
+llama_perf_context_print: prompt eval time =  154546.03 ms / 36666 tokens (    4.21 ms per token,   237.25 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  155368.54 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf (version GGUF V3 (latest))
+
+750	79.86666667%	[76.8479%, 82.5810%]
+
+
+llama_perf_context_print:        load time =     746.08 ms
+llama_perf_context_print: prompt eval time =  544955.95 ms / 129319 tokens (    4.21 ms per token,   237.30 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  548715.91 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf (version GGUF V3 (latest))
+
+Final result: 42.0000 +/- 1.8034
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     771.21 ms
+llama_perf_context_print: prompt eval time =  299726.08 ms / 68956 tokens (    4.35 ms per token,   230.06 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  301016.69 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  21.165413 ±   0.164512
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  73.80%
+Mean ln(PPL(Q)/PPL(base))     :   1.230957 ±   0.005322
+Mean PPL(Q)/PPL(base)         :   3.424504 ±   0.018226
+Mean PPL(Q)-PPL(base)         :  14.984836 ±   0.137053
+
+====== KL divergence statistics ======
+Mean    KLD:   1.340446 ±   0.004301
+Maximum KLD:  26.031479
+99.9%   KLD:  13.794025
+99.0%   KLD:   8.598367
+99.0%   KLD:   8.598367
+Median  KLD:   0.843369
+10.0%   KLD:   0.087419
+ 5.0%   KLD:   0.032046
+ 1.0%   KLD:   0.005851
+Minimum KLD:   0.000204
+
+====== Token probability statistics ======
+Mean    Δp: -19.454 ± 0.086 %
+Maximum Δp: 95.317%
+99.9%   Δp: 71.779%
+99.0%   Δp: 46.392%
+95.0%   Δp: 20.287%
+90.0%   Δp:  7.866%
+75.0%   Δp: -0.004%
+Median  Δp: -5.901%
+25.0%   Δp: -35.629%
+10.0%   Δp: -76.898%
+ 5.0%   Δp: -92.591%
+ 1.0%   Δp: -99.784%
+ 0.1%   Δp: -99.968%
+Minimum Δp: -99.998%
+RMS Δp    : 38.586 ± 0.088 %
+Same top p: 61.705 ± 0.125 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf (version GGUF V3 (latest))
+
+Final result: 38.0000 +/- 1.7736
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     775.02 ms
+llama_perf_context_print: prompt eval time =  229109.68 ms / 51053 tokens (    4.49 ms per token,   222.83 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  230721.17 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq3_s.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ3_S.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 72.5333 +/- 1.6309
+
+llama_perf_context_print:        load time =     766.80 ms
+llama_perf_context_print: prompt eval time =   99506.70 ms / 22541 tokens (    4.41 ms per token,   226.53 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =   99996.10 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf (version GGUF V3 (latest))
+
+Final result: 68.4000 +/- 1.6988
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    5388.44 ms
+llama_perf_context_print: prompt eval time =  152847.24 ms / 36666 tokens (    4.17 ms per token,   239.89 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  153655.01 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf (version GGUF V3 (latest))
+
+750	80.66666667%	[77.6870%, 83.3338%]
+
+
+llama_perf_context_print:        load time =     881.76 ms
+llama_perf_context_print: prompt eval time =  557982.84 ms / 129319 tokens (    4.31 ms per token,   231.76 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  561719.88 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf (version GGUF V3 (latest))
+
+Final result: 44.9333 +/- 1.8176
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     922.07 ms
+llama_perf_context_print: prompt eval time =  295625.34 ms / 68956 tokens (    4.29 ms per token,   233.25 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  296873.16 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.783744 ±   0.146959
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  74.79%
+Mean ln(PPL(Q)/PPL(base))     :   1.111580 ±   0.005253
+Mean PPL(Q)/PPL(base)         :   3.039157 ±   0.015966
+Mean PPL(Q)-PPL(base)         :  12.603167 ±   0.119417
+
+====== KL divergence statistics ======
+Mean    KLD:   1.199318 ±   0.004258
+Maximum KLD:  26.543749
+99.9%   KLD:  14.340773
+99.0%   KLD:   8.742259
+99.0%   KLD:   8.742259
+Median  KLD:   0.715601
+10.0%   KLD:   0.071172
+ 5.0%   KLD:   0.025864
+ 1.0%   KLD:   0.004589
+Minimum KLD:   0.000142
+
+====== Token probability statistics ======
+Mean    Δp: -17.171 ± 0.083 %
+Maximum Δp: 92.904%
+99.9%   Δp: 72.496%
+99.0%   Δp: 48.216%
+95.0%   Δp: 22.569%
+90.0%   Δp: 10.000%
+75.0%   Δp:  0.011%
+Median  Δp: -4.438%
+25.0%   Δp: -30.853%
+10.0%   Δp: -71.739%
+ 5.0%   Δp: -90.613%
+ 1.0%   Δp: -99.767%
+ 0.1%   Δp: -99.967%
+Minimum Δp: -99.998%
+RMS Δp    : 36.745 ± 0.088 %
+Same top p: 64.223 ± 0.123 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf (version GGUF V3 (latest))
+
+Final result: 38.1333 +/- 1.7748
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     920.15 ms
+llama_perf_context_print: prompt eval time =  224775.45 ms / 51053 tokens (    4.40 ms per token,   227.13 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  226389.03 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-iq4_nl.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-IQ4_NL.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 74.4000 +/- 1.5947
+
+llama_perf_context_print:        load time =     863.03 ms
+llama_perf_context_print: prompt eval time =   98796.90 ms / 22541 tokens (    4.38 ms per token,   228.15 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =   99295.86 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf (version GGUF V3 (latest))
+
+Final result: 67.2000 +/- 1.7155
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    4789.37 ms
+llama_perf_context_print: prompt eval time =  171359.80 ms / 36666 tokens (    4.67 ms per token,   213.97 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  172182.70 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf (version GGUF V3 (latest))
+
+750	80.26666667%	[77.2672%, 82.9576%]
+
+
+llama_perf_context_print:        load time =     819.35 ms
+llama_perf_context_print: prompt eval time =  605874.35 ms / 129319 tokens (    4.69 ms per token,   213.44 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  609526.15 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf (version GGUF V3 (latest))
+
+Final result: 43.2000 +/- 1.8100
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     873.17 ms
+llama_perf_context_print: prompt eval time =  318870.48 ms / 68956 tokens (    4.62 ms per token,   216.25 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  320164.35 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  19.313300 ±   0.150799
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  74.61%
+Mean ln(PPL(Q)/PPL(base))     :   1.139382 ±   0.005262
+Mean PPL(Q)/PPL(base)         :   3.124838 ±   0.016443
+Mean PPL(Q)-PPL(base)         :  13.132723 ±   0.123247
+
+====== KL divergence statistics ======
+Mean    KLD:   1.248712 ±   0.004216
+Maximum KLD:  28.765745
+99.9%   KLD:  13.682988
+99.0%   KLD:   8.611128
+99.0%   KLD:   8.611128
+Median  KLD:   0.769048
+10.0%   KLD:   0.075574
+ 5.0%   KLD:   0.027425
+ 1.0%   KLD:   0.004777
+Minimum KLD:   0.000121
+
+====== Token probability statistics ======
+Mean    Δp: -17.672 ± 0.084 %
+Maximum Δp: 94.089%
+99.9%   Δp: 73.751%
+99.0%   Δp: 50.009%
+95.0%   Δp: 22.814%
+90.0%   Δp:  9.454%
+75.0%   Δp:  0.004%
+Median  Δp: -4.889%
+25.0%   Δp: -32.188%
+10.0%   Δp: -72.799%
+ 5.0%   Δp: -91.232%
+ 1.0%   Δp: -99.761%
+ 0.1%   Δp: -99.966%
+Minimum Δp: -99.998%
+RMS Δp    : 37.260 ± 0.088 %
+Same top p: 62.749 ± 0.124 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf (version GGUF V3 (latest))
+
+Final result: 39.6000 +/- 1.7870
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     858.54 ms
+llama_perf_context_print: prompt eval time =  243423.07 ms / 51053 tokens (    4.77 ms per token,   209.73 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  245001.45 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_l.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_L.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 72.9333 +/- 1.6235
+
+llama_perf_context_print:        load time =     903.93 ms
+llama_perf_context_print: prompt eval time =  106212.48 ms / 22541 tokens (    4.71 ms per token,   212.23 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  106738.43 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 66.6667 +/- 1.7225
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    4267.81 ms
+llama_perf_context_print: prompt eval time =  165656.48 ms / 36666 tokens (    4.52 ms per token,   221.34 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  166474.66 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf (version GGUF V3 (latest))
+
+750	80.66666667%	[77.6870%, 83.3338%]
+
+
+llama_perf_context_print:        load time =     756.25 ms
+llama_perf_context_print: prompt eval time =  584702.34 ms / 129319 tokens (    4.52 ms per token,   221.17 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  588407.02 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 43.8667 +/- 1.8132
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     809.88 ms
+llama_perf_context_print: prompt eval time =  300457.60 ms / 68956 tokens (    4.36 ms per token,   229.50 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  301704.70 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.723777 ±   0.145380
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  75.90%
+Mean ln(PPL(Q)/PPL(base))     :   1.108382 ±   0.005110
+Mean PPL(Q)/PPL(base)         :   3.029454 ±   0.015481
+Mean PPL(Q)-PPL(base)         :  12.543199 ±   0.117315
+
+====== KL divergence statistics ======
+Mean    KLD:   1.226150 ±   0.004006
+Maximum KLD:  27.303829
+99.9%   KLD:  13.319038
+99.0%   KLD:   8.045850
+99.0%   KLD:   8.045850
+Median  KLD:   0.778573
+10.0%   KLD:   0.072866
+ 5.0%   KLD:   0.026539
+ 1.0%   KLD:   0.004645
+Minimum KLD:   0.000149
+
+====== Token probability statistics ======
+Mean    Δp: -17.217 ± 0.084 %
+Maximum Δp: 93.510%
+99.9%   Δp: 74.449%
+99.0%   Δp: 50.740%
+95.0%   Δp: 23.161%
+90.0%   Δp:  9.707%
+75.0%   Δp:  0.007%
+Median  Δp: -4.675%
+25.0%   Δp: -31.413%
+10.0%   Δp: -71.201%
+ 5.0%   Δp: -90.111%
+ 1.0%   Δp: -99.699%
+ 0.1%   Δp: -99.959%
+Minimum Δp: -99.998%
+RMS Δp    : 36.807 ± 0.087 %
+Same top p: 63.002 ± 0.124 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 39.4667 +/- 1.7860
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     794.64 ms
+llama_perf_context_print: prompt eval time =  226374.98 ms / 51053 tokens (    4.43 ms per token,   225.52 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  228005.49 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_m.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_M.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 72.2667 +/- 1.6358
+
+llama_perf_context_print:        load time =     786.85 ms
+llama_perf_context_print: prompt eval time =   98570.95 ms / 22541 tokens (    4.37 ms per token,   228.68 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =   99099.17 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 66.2667 +/- 1.7276
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    4024.90 ms
+llama_perf_context_print: prompt eval time =  162864.58 ms / 36666 tokens (    4.44 ms per token,   225.13 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  163701.55 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf (version GGUF V3 (latest))
+
+750	78.93333333%	[75.8712%, 81.7006%]
+
+
+llama_perf_context_print:        load time =     717.95 ms
+llama_perf_context_print: prompt eval time =  573377.50 ms / 129319 tokens (    4.43 ms per token,   225.54 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  577152.30 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 43.7333 +/- 1.8126
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     729.81 ms
+llama_perf_context_print: prompt eval time =  302270.16 ms / 68956 tokens (    4.38 ms per token,   228.13 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  303540.55 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  19.765437 ±   0.153182
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  74.13%
+Mean ln(PPL(Q)/PPL(base))     :   1.162523 ±   0.005278
+Mean PPL(Q)/PPL(base)         :   3.197992 ±   0.016878
+Mean PPL(Q)-PPL(base)         :  13.584860 ±   0.125811
+
+====== KL divergence statistics ======
+Mean    KLD:   1.295119 ±   0.004177
+Maximum KLD:  27.306818
+99.9%   KLD:  13.228414
+99.0%   KLD:   8.401047
+99.0%   KLD:   8.401047
+Median  KLD:   0.824191
+10.0%   KLD:   0.081751
+ 5.0%   KLD:   0.029987
+ 1.0%   KLD:   0.004980
+Minimum KLD:   0.000157
+
+====== Token probability statistics ======
+Mean    Δp: -18.417 ± 0.085 %
+Maximum Δp: 92.683%
+99.9%   Δp: 74.856%
+99.0%   Δp: 50.844%
+95.0%   Δp: 22.899%
+90.0%   Δp:  9.183%
+75.0%   Δp:  0.001%
+Median  Δp: -5.493%
+25.0%   Δp: -34.119%
+10.0%   Δp: -74.211%
+ 5.0%   Δp: -91.714%
+ 1.0%   Δp: -99.759%
+ 0.1%   Δp: -99.964%
+Minimum Δp: -99.995%
+RMS Δp    : 38.004 ± 0.087 %
+Same top p: 61.136 ± 0.125 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 38.1333 +/- 1.7748
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     713.93 ms
+llama_perf_context_print: prompt eval time =  230959.05 ms / 51053 tokens (    4.52 ms per token,   221.05 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  232565.83 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q3_k_s.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q3_K_S.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 72.8000 +/- 1.6260
+
+llama_perf_context_print:        load time =     732.76 ms
+llama_perf_context_print: prompt eval time =  100522.70 ms / 22541 tokens (    4.46 ms per token,   224.24 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  101025.93 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 68.0000 +/- 1.7045
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    5558.39 ms
+llama_perf_context_print: prompt eval time =  161414.80 ms / 36666 tokens (    4.40 ms per token,   227.15 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  162273.65 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf (version GGUF V3 (latest))
+
+750	80.93333333%	[77.9671%, 83.5843%]
+
+
+llama_perf_context_print:        load time =     911.43 ms
+llama_perf_context_print: prompt eval time =  570839.24 ms / 129319 tokens (    4.41 ms per token,   226.54 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  574520.13 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 45.2000 +/- 1.8185
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     974.87 ms
+llama_perf_context_print: prompt eval time =  300441.31 ms / 68956 tokens (    4.36 ms per token,   229.52 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  301727.28 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.556910 ±   0.145472
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  74.92%
+Mean ln(PPL(Q)/PPL(base))     :   1.099431 ±   0.005248
+Mean PPL(Q)/PPL(base)         :   3.002456 ±   0.015757
+Mean PPL(Q)-PPL(base)         :  12.376333 ±   0.117900
+
+====== KL divergence statistics ======
+Mean    KLD:   1.187728 ±   0.004237
+Maximum KLD:  27.586824
+99.9%   KLD:  14.366529
+99.0%   KLD:   8.795478
+99.0%   KLD:   8.795478
+Median  KLD:   0.711806
+10.0%   KLD:   0.068614
+ 5.0%   KLD:   0.024978
+ 1.0%   KLD:   0.004366
+Minimum KLD:   0.000131
+
+====== Token probability statistics ======
+Mean    Δp: -16.803 ± 0.083 %
+Maximum Δp: 94.144%
+99.9%   Δp: 72.562%
+99.0%   Δp: 49.033%
+95.0%   Δp: 23.292%
+90.0%   Δp: 10.354%
+75.0%   Δp:  0.021%
+Median  Δp: -4.202%
+25.0%   Δp: -30.211%
+10.0%   Δp: -70.964%
+ 5.0%   Δp: -90.473%
+ 1.0%   Δp: -99.772%
+ 0.1%   Δp: -99.967%
+Minimum Δp: -99.998%
+RMS Δp    : 36.521 ± 0.088 %
+Same top p: 63.986 ± 0.123 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 36.6667 +/- 1.7608
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     953.61 ms
+llama_perf_context_print: prompt eval time =  229781.03 ms / 51053 tokens (    4.50 ms per token,   222.18 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  231385.66 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_m.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_M.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 72.1333 +/- 1.6382
+
+llama_perf_context_print:        load time =     981.15 ms
+llama_perf_context_print: prompt eval time =  100004.38 ms / 22541 tokens (    4.44 ms per token,   225.40 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  100554.46 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 67.0667 +/- 1.7172
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    5272.61 ms
+llama_perf_context_print: prompt eval time =  161319.58 ms / 36666 tokens (    4.40 ms per token,   227.29 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  162146.27 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf (version GGUF V3 (latest))
+
+750	81.06666667%	[78.1072%, 83.7095%]
+
+
+llama_perf_context_print:        load time =     868.43 ms
+llama_perf_context_print: prompt eval time =  569329.51 ms / 129319 tokens (    4.40 ms per token,   227.14 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  573001.51 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 45.2000 +/- 1.8185
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =     885.80 ms
+llama_perf_context_print: prompt eval time =  300092.31 ms / 68956 tokens (    4.35 ms per token,   229.78 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  301434.44 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.663517 ±   0.146425
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  74.87%
+Mean ln(PPL(Q)/PPL(base))     :   1.105159 ±   0.005257
+Mean PPL(Q)/PPL(base)         :   3.019704 ±   0.015873
+Mean PPL(Q)-PPL(base)         :  12.482940 ±   0.118853
+
+====== KL divergence statistics ======
+Mean    KLD:   1.192878 ±   0.004250
+Maximum KLD:  27.191839
+99.9%   KLD:  14.356927
+99.0%   KLD:   8.760485
+99.0%   KLD:   8.760485
+Median  KLD:   0.713994
+10.0%   KLD:   0.069566
+ 5.0%   KLD:   0.025085
+ 1.0%   KLD:   0.004382
+Minimum KLD:   0.000130
+
+====== Token probability statistics ======
+Mean    Δp: -16.887 ± 0.083 %
+Maximum Δp: 94.834%
+99.9%   Δp: 72.359%
+99.0%   Δp: 48.972%
+95.0%   Δp: 23.124%
+90.0%   Δp: 10.343%
+75.0%   Δp:  0.019%
+Median  Δp: -4.257%
+25.0%   Δp: -30.401%
+10.0%   Δp: -71.279%
+ 5.0%   Δp: -90.560%
+ 1.0%   Δp: -99.771%
+ 0.1%   Δp: -99.966%
+Minimum Δp: -99.998%
+RMS Δp    : 36.598 ± 0.088 %
+Same top p: 63.979 ± 0.123 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 36.2667 +/- 1.7567
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     868.82 ms
+llama_perf_context_print: prompt eval time =  229327.93 ms / 51053 tokens (    4.49 ms per token,   222.62 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  230969.11 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q4_k_s.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q4_K_S.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 72.0000 +/- 1.6406
+
+llama_perf_context_print:        load time =     885.47 ms
+llama_perf_context_print: prompt eval time =   99462.32 ms / 22541 tokens (    4.41 ms per token,   226.63 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =   99973.97 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 67.0667 +/- 1.7172
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    6034.32 ms
+llama_perf_context_print: prompt eval time =  171299.05 ms / 36666 tokens (    4.67 ms per token,   214.05 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  172122.97 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf (version GGUF V3 (latest))
+
+750	81.73333333%	[78.8086%, 84.3346%]
+
+
+llama_perf_context_print:        load time =    1034.44 ms
+llama_perf_context_print: prompt eval time =  604834.04 ms / 129319 tokens (    4.68 ms per token,   213.81 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  608474.71 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 44.5333 +/- 1.8160
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =    1042.58 ms
+llama_perf_context_print: prompt eval time =  317286.78 ms / 68956 tokens (    4.60 ms per token,   217.33 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  318545.57 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.174846 ±   0.142320
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  75.14%
+Mean ln(PPL(Q)/PPL(base))     :   1.078627 ±   0.005222
+Mean PPL(Q)/PPL(base)         :   2.940639 ±   0.015356
+Mean PPL(Q)-PPL(base)         :  11.994269 ±   0.114726
+
+====== KL divergence statistics ======
+Mean    KLD:   1.159685 ±   0.004238
+Maximum KLD:  28.100733
+99.9%   KLD:  14.541190
+99.0%   KLD:   8.790474
+99.0%   KLD:   8.790474
+Median  KLD:   0.682733
+10.0%   KLD:   0.066346
+ 5.0%   KLD:   0.024022
+ 1.0%   KLD:   0.004240
+Minimum KLD:   0.000159
+
+====== Token probability statistics ======
+Mean    Δp: -16.552 ± 0.083 %
+Maximum Δp: 94.307%
+99.9%   Δp: 72.111%
+99.0%   Δp: 47.813%
+95.0%   Δp: 23.079%
+90.0%   Δp: 10.497%
+75.0%   Δp:  0.030%
+Median  Δp: -4.029%
+25.0%   Δp: -29.474%
+10.0%   Δp: -70.364%
+ 5.0%   Δp: -90.072%
+ 1.0%   Δp: -99.761%
+ 0.1%   Δp: -99.966%
+Minimum Δp: -99.998%
+RMS Δp    : 36.214 ± 0.088 %
+Same top p: 64.455 ± 0.123 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf (version GGUF V3 (latest))
+
+Final result: 37.8667 +/- 1.7724
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =    1018.74 ms
+llama_perf_context_print: prompt eval time =  243028.85 ms / 51053 tokens (    4.76 ms per token,   210.07 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  244621.86 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_m.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_M.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 73.8667 +/- 1.6054
+
+llama_perf_context_print:        load time =    1080.02 ms
+llama_perf_context_print: prompt eval time =  105347.75 ms / 22541 tokens (    4.67 ms per token,   213.97 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  105836.33 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 67.3333 +/- 1.7137
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    5824.40 ms
+llama_perf_context_print: prompt eval time =  172340.28 ms / 36666 tokens (    4.70 ms per token,   212.75 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  173198.86 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf (version GGUF V3 (latest))
+
+750	81.46666667%	[78.5279%, 84.0847%]
+
+
+llama_perf_context_print:        load time =    1049.43 ms
+llama_perf_context_print: prompt eval time =  592449.23 ms / 129319 tokens (    4.58 ms per token,   218.28 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  596100.86 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 44.2667 +/- 1.8149
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =    1030.41 ms
+llama_perf_context_print: prompt eval time =  307801.57 ms / 68956 tokens (    4.46 ms per token,   224.03 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  309115.30 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.199918 ±   0.142513
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  75.20%
+Mean ln(PPL(Q)/PPL(base))     :   1.080005 ±   0.005216
+Mean PPL(Q)/PPL(base)         :   2.944695 ±   0.015358
+Mean PPL(Q)-PPL(base)         :  12.019340 ±   0.114880
+
+====== KL divergence statistics ======
+Mean    KLD:   1.160040 ±   0.004229
+Maximum KLD:  27.444889
+99.9%   KLD:  14.604442
+99.0%   KLD:   8.763094
+99.0%   KLD:   8.763094
+Median  KLD:   0.682655
+10.0%   KLD:   0.066186
+ 5.0%   KLD:   0.023837
+ 1.0%   KLD:   0.004326
+Minimum KLD:   0.000148
+
+====== Token probability statistics ======
+Mean    Δp: -16.565 ± 0.083 %
+Maximum Δp: 94.238%
+99.9%   Δp: 72.114%
+99.0%   Δp: 47.597%
+95.0%   Δp: 23.064%
+90.0%   Δp: 10.544%
+75.0%   Δp:  0.030%
+Median  Δp: -4.020%
+25.0%   Δp: -29.497%
+10.0%   Δp: -70.424%
+ 5.0%   Δp: -90.053%
+ 1.0%   Δp: -99.759%
+ 0.1%   Δp: -99.965%
+Minimum Δp: -99.998%
+RMS Δp    : 36.220 ± 0.088 %
+Same top p: 64.431 ± 0.123 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf (version GGUF V3 (latest))
+
+Final result: 38.6667 +/- 1.7794
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =     984.39 ms
+llama_perf_context_print: prompt eval time =  235496.75 ms / 51053 tokens (    4.61 ms per token,   216.79 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  237131.13 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q5_k_s.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q5_K_S.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 74.2667 +/- 1.5974
+
+llama_perf_context_print:        load time =    1051.46 ms
+llama_perf_context_print: prompt eval time =  102348.25 ms / 22541 tokens (    4.54 ms per token,   220.24 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  102869.23 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf (version GGUF V3 (latest))
+
+Final result: 67.4667 +/- 1.7119
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =    7664.20 ms
+llama_perf_context_print: prompt eval time =  175407.42 ms / 36666 tokens (    4.78 ms per token,   209.03 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  176202.75 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf (version GGUF V3 (latest))
+
+750	81.06666667%	[78.1072%, 83.7095%]
+
+
+llama_perf_context_print:        load time =    1225.60 ms
+llama_perf_context_print: prompt eval time =  622423.40 ms / 129319 tokens (    4.81 ms per token,   207.77 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  626130.15 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf (version GGUF V3 (latest))
+
+Final result: 44.5333 +/- 1.8160
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =    1255.72 ms
+llama_perf_context_print: prompt eval time =  327772.92 ms / 68956 tokens (    4.75 ms per token,   210.38 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  329047.99 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.213825 ±   0.142965
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  75.05%
+Mean ln(PPL(Q)/PPL(base))     :   1.080769 ±   0.005241
+Mean PPL(Q)/PPL(base)         :   2.946946 ±   0.015446
+Mean PPL(Q)-PPL(base)         :  12.033248 ±   0.115399
+
+====== KL divergence statistics ======
+Mean    KLD:   1.158026 ±   0.004262
+Maximum KLD:  26.265640
+99.9%   KLD:  14.634221
+99.0%   KLD:   8.878881
+99.0%   KLD:   8.878881
+Median  KLD:   0.679157
+10.0%   KLD:   0.065587
+ 5.0%   KLD:   0.023885
+ 1.0%   KLD:   0.004187
+Minimum KLD:   0.000139
+
+====== Token probability statistics ======
+Mean    Δp: -16.509 ± 0.083 %
+Maximum Δp: 94.239%
+99.9%   Δp: 72.141%
+99.0%   Δp: 48.282%
+95.0%   Δp: 23.325%
+90.0%   Δp: 10.605%
+75.0%   Δp:  0.032%
+Median  Δp: -4.000%
+25.0%   Δp: -29.434%
+10.0%   Δp: -70.325%
+ 5.0%   Δp: -90.061%
+ 1.0%   Δp: -99.771%
+ 0.1%   Δp: -99.967%
+Minimum Δp: -99.998%
+RMS Δp    : 36.219 ± 0.088 %
+Same top p: 64.447 ± 0.123 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf (version GGUF V3 (latest))
+
+Final result: 39.6000 +/- 1.7870
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =    1251.53 ms
+llama_perf_context_print: prompt eval time =  250368.11 ms / 51053 tokens (    4.90 ms per token,   203.91 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  251979.66 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q6_k.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q6_K.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 73.8667 +/- 1.6054
+
+llama_perf_context_print:        load time =    1250.90 ms
+llama_perf_context_print: prompt eval time =  108113.49 ms / 22541 tokens (    4.80 ms per token,   208.49 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  108625.99 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.arc
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.arc
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf (version GGUF V3 (latest))
+
+Final result: 68.1333 +/- 1.7026
+Random chance: 25.0083 +/- 1.5824
+
+
+llama_perf_context_print:        load time =   10486.40 ms
+llama_perf_context_print: prompt eval time =  161243.77 ms / 36666 tokens (    4.40 ms per token,   227.39 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  162055.73 ms / 36667 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.hsw
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.hsw
@@ -0,0 +1,20 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf (version GGUF V3 (latest))
+
+750	81.33333333%	[78.3876%, 83.9597%]
+
+
+llama_perf_context_print:        load time =    1550.41 ms
+llama_perf_context_print: prompt eval time =  553902.33 ms / 129319 tokens (    4.28 ms per token,   233.47 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  557360.27 ms / 129320 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.mmlu
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.mmlu
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf (version GGUF V3 (latest))
+
+Final result: 44.9333 +/- 1.8176
+Random chance: 25.0000 +/- 1.5822
+
+
+llama_perf_context_print:        load time =    1526.16 ms
+llama_perf_context_print: prompt eval time =  290805.14 ms / 68956 tokens (    4.22 ms per token,   237.12 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  292006.82 ms / 68957 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.ppx
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.ppx
@@ -0,0 +1,37 @@
+====== Perplexity statistics ======
+Mean PPL(Q)                   :  18.203515 ±   0.142826
+Mean PPL(base)                :   6.180577 ±   0.041038
+Cor(ln(PPL(Q)), ln(PPL(base))):  75.02%
+Mean ln(PPL(Q)/PPL(base))     :   1.080203 ±   0.005242
+Mean PPL(Q)/PPL(base)         :   2.945277 ±   0.015439
+Mean PPL(Q)-PPL(base)         :  12.022938 ±   0.115276
+
+====== KL divergence statistics ======
+Mean    KLD:   1.158351 ±   0.004265
+Maximum KLD:  27.082415
+99.9%   KLD:  14.510898
+99.0%   KLD:   8.873251
+99.0%   KLD:   8.873251
+Median  KLD:   0.678687
+10.0%   KLD:   0.065954
+ 5.0%   KLD:   0.024006
+ 1.0%   KLD:   0.004161
+Minimum KLD:   0.000141
+
+====== Token probability statistics ======
+Mean    Δp: -16.515 ± 0.083 %
+Maximum Δp: 94.585%
+99.9%   Δp: 72.074%
+99.0%   Δp: 48.296%
+95.0%   Δp: 23.169%
+90.0%   Δp: 10.662%
+75.0%   Δp:  0.031%
+Median  Δp: -4.006%
+25.0%   Δp: -29.412%
+10.0%   Δp: -70.362%
+ 5.0%   Δp: -90.143%
+ 1.0%   Δp: -99.767%
+ 0.1%   Δp: -99.966%
+Minimum Δp: -99.998%
+RMS Δp    : 36.227 ± 0.088 %
+Same top p: 64.562 ± 0.123 %
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.tqa
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.tqa
@@ -0,0 +1,21 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf (version GGUF V3 (latest))
+
+Final result: 38.2667 +/- 1.7759
+Random chance: 19.8992 +/- 1.4588
+
+
+llama_perf_context_print:        load time =    1498.38 ms
+llama_perf_context_print: prompt eval time =  227802.62 ms / 51053 tokens (    4.46 ms per token,   224.11 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  229374.00 ms / 51054 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
--- a/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.wng
+++ b/scores/Dolphin-Mistral-24B-Venice-Edition-q8_0.wng
@@ -0,0 +1,19 @@
+build: 5770 (b25e9277) with Apple clang version 17.0.0 (clang-1700.0.13.3) for arm64-apple-darwin24.4.0
+llama_model_load_from_file_impl: using device Metal (Apple M4 Max) - 49151 MiB free
+llama_model_loader: loaded meta data with 43 key-value pairs and 345 tensors from ./Dolphin-Mistral-24B-Venice-Edition-pruned-Q8_0.gguf (version GGUF V3 (latest))
+
+Final Winogrande score(750 tasks): 74.4000 +/- 1.5947
+
+llama_perf_context_print:        load time =    1463.29 ms
+llama_perf_context_print: prompt eval time =  100036.10 ms / 22541 tokens (    4.44 ms per token,   225.33 tokens per second)
+llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
+llama_perf_context_print:       total time =  100546.29 ms / 22542 tokens
+ggml_metal_free: deallocating
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
+ggml_metal_mem_pool_free: freeing memory pool, num heaps = 0 (total = 0)
Author	SHA1	Message	Date
Ed Addario	c2d8908143	Update README.md	2025-07-01 09:07:31 +01:00
Ed Addario	fde40b8129	Update README.md	2025-07-01 08:59:38 +01:00
Ed Addario	35b1741d2b	Add GGUF internal file structure	2025-07-01 08:19:46 +01:00
Ed Addario	664228d4d7	Generate Perplexity, KLD, ARC, HellaSwag, MMLU, Truthful QA and WinoGrande scores	2025-07-01 08:18:25 +01:00
Ed Addario	3a6bd3ae0d	Layer-wise & pruned quantization IQ3_M	2025-07-01 07:38:42 +01:00
Ed Addario	6e9dd7f6ee	Layer-wise & pruned quantization IQ3_S	2025-07-01 07:31:37 +01:00
Ed Addario	7860f6907e	Layer-wise & pruned quantization IQ4_NL	2025-07-01 06:53:26 +01:00
Ed Addario	dae2ab35db	Layer-wise & pruned quantization Q3_K_L	2025-07-01 06:35:54 +01:00
Ed Addario	36ef14d1fd	Layer-wise & pruned quantization Q3_K_M	2025-06-30 23:02:04 +01:00
Ed Addario	0669ee1ce7	Layer-wise & pruned quantization Q3_K_S	2025-06-30 22:54:54 +01:00