Downloads · 30 days
430
6% of all-time downloads
RichardErkhov/maldv_-_winter-garden-7b-delta-gguf
maldv_-_winter-garden-7b-delta-gguf is a machine learning model from RichardErkhov. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Downloads · 30 days
430
6% of all-time downloads
All-time downloads
6.7K
Public
Repo size
82.2 GB
Likes
0
Public
Click a slice to open those files.
.gguf95.2 GB · 100%
From the Hugging Face model README
Quantization made by Richard Erkhov.
winter-garden-7b-delta - GGUF
| Name | Quant method | Size |
|---|---|---|
| winter-garden-7b-delta.Q2_K.gguf | Q2_K | 2.53GB |
| winter-garden-7b-delta.IQ3_XS.gguf | IQ3_XS | 2.81GB |
| winter-garden-7b-delta.IQ3_S.gguf | IQ3_S | 2.96GB |
| winter-garden-7b-delta.Q3_K_S.gguf | Q3_K_S | 2.95GB |
| winter-garden-7b-delta.IQ3_M.gguf | IQ3_M | 3.06GB |
| winter-garden-7b-delta.Q3_K.gguf | Q3_K | 3.28GB |
| winter-garden-7b-delta.Q3_K_M.gguf | Q3_K_M | 3.28GB |
| winter-garden-7b-delta.Q3_K_L.gguf | Q3_K_L | 3.56GB |
| winter-garden-7b-delta.IQ4_XS.gguf | IQ4_XS | 3.67GB |
| winter-garden-7b-delta.Q4_0.gguf | Q4_0 | 3.83GB |
| winter-garden-7b-delta.IQ4_NL.gguf | IQ4_NL | 3.87GB |
| winter-garden-7b-delta.Q4_K_S.gguf | Q4_K_S | 3.86GB |
| winter-garden-7b-delta.Q4_K.gguf | Q4_K | 4.07GB |
| winter-garden-7b-delta.Q4_K_M.gguf | Q4_K_M | 4.07GB |
| winter-garden-7b-delta.Q4_1.gguf | Q4_1 | 4.24GB |
| winter-garden-7b-delta.Q5_0.gguf | Q5_0 | 4.65GB |
| winter-garden-7b-delta.Q5_K_S.gguf | Q5_K_S | 4.65GB |
| winter-garden-7b-delta.Q5_K.gguf | Q5_K | 4.78GB |
| winter-garden-7b-delta.Q5_K_M.gguf | Q5_K_M | 4.78GB |
| winter-garden-7b-delta.Q5_1.gguf | Q5_1 | 5.07GB |
| winter-garden-7b-delta.Q6_K.gguf | Q6_K | 5.53GB |
| winter-garden-7b-delta.Q8_0.gguf | Q8_0 | 7.17GB |
license: cc-by-nc-4.0 tags:
It was mentioned that we are in the open ai dark winter; so I thought I would make myself a nice winter garden.
I performed the same type of merge as in the previous model, but with a different set of models. I took the following models:
and merged in
in an iterative DARE-TIES tree merge, ordering the merge order by tensor-relative cosine similarity until the merge branches resolve to a single value.
These models were selected because they follow my chat template, which is '</s>' ended turns. A lot of models follow this template by default because they were trained with end padding, so this is a natural choice for chat, and should be highly compatible with ST.
Tom: Hello, how are you?</s>
Jane: I am fine, thank you.</s>
The purpose of all of these models is to act as a base for me to train on. This one so far has the best multi-turn conversational ability, and should get really good at following long-form conversations after a bit of tweaking.
| Metric | Score |
|---|---|
| Average | 64.93 |
| ARC | 64.16 |
| HellaSwag | 84.37 |
| MMLU | 60.38 |
| TruthfulQA | 67.95 |
| Winogrande | 76.72 |
| GSM8K | 36.01 |