Upload folder using huggingface_hub
Browse files
README.md
CHANGED
@@ -8,7 +8,7 @@ tags:
|
|
8 |
- merge
|
9 |
|
10 |
---
|
11 |
-
#
|
12 |
|
13 |
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
14 |
|
@@ -21,94 +21,7 @@ This model was merged using the [TIES](https://arxiv.org/abs/2306.01708) merge m
|
|
21 |
|
22 |
The following models were included in the merge:
|
23 |
* [unsloth/Qwen2.5-3B-Instruct](https://huggingface.co/unsloth/Qwen2.5-3B-Instruct)
|
24 |
-
* triples
|
25 |
-
* genstruct
|
26 |
-
* kg
|
27 |
|
28 |
-
### Configuration
|
29 |
-
|
30 |
-
The following YAML configuration was used to produce this model:
|
31 |
-
|
32 |
-
```yaml
|
33 |
-
models:
|
34 |
-
# Base instructed model
|
35 |
-
- model: unsloth/Qwen2.5-3B-Instruct
|
36 |
-
parameters:
|
37 |
-
weight: 1
|
38 |
-
density: 1
|
39 |
-
|
40 |
-
# Merged LoRA models
|
41 |
-
- model: genstruct/merged_model
|
42 |
-
parameters:
|
43 |
-
weight: 1.0
|
44 |
-
density: 1.0
|
45 |
-
|
46 |
-
# - model: summary/merged_model
|
47 |
-
# parameters:
|
48 |
-
# weight: 1.0
|
49 |
-
# density: 1.0
|
50 |
-
|
51 |
-
- model: kg/merged_model
|
52 |
-
parameters:
|
53 |
-
weight: 1.0
|
54 |
-
density: 1.0
|
55 |
-
|
56 |
-
#### THIS BREAKS KG!!!
|
57 |
-
# - model: pII/merged_model
|
58 |
-
# parameters:
|
59 |
-
# weight: 1.0
|
60 |
-
# density: 1.0
|
61 |
-
|
62 |
-
# #### Breaks KG!
|
63 |
-
# - model: preference/merged_model
|
64 |
-
# parameters:
|
65 |
-
# weight: 1.0
|
66 |
-
# density: 1.0
|
67 |
-
|
68 |
-
- model: triples/merged_model
|
69 |
-
parameters:
|
70 |
-
weight: 1.0
|
71 |
-
density: 1.0
|
72 |
-
|
73 |
-
# - model: suitable/merged_model
|
74 |
-
# parameters:
|
75 |
-
# weight: 1.0
|
76 |
-
# density: 1.0
|
77 |
-
|
78 |
-
# - model: feedback/merged_model
|
79 |
-
# parameters:
|
80 |
-
# weight: 1.0
|
81 |
-
# density: 1.0
|
82 |
-
|
83 |
-
# Merge configuration
|
84 |
-
merge_method: ties
|
85 |
-
base_model: unsloth/Qwen2.5-3B
|
86 |
-
parameters:
|
87 |
-
normalize: true
|
88 |
-
int8_mask: true
|
89 |
-
dtype: bfloat16
|
90 |
-
|
91 |
-
# # Tokenizer configuration
|
92 |
-
# tokenizer_source: Qwen/Qwen1.5-14B-Chat
|
93 |
-
# tokenizer_parameters:
|
94 |
-
# trust_remote_code: true
|
95 |
-
|
96 |
-
# # Output configuration
|
97 |
-
# output:
|
98 |
-
# precision: bfloat16
|
99 |
-
# model_format: safetensors
|
100 |
-
# max_shard_size: "4GB"
|
101 |
-
|
102 |
-
# # Training configuration (for potential fine-tuning)
|
103 |
-
# training:
|
104 |
-
# learning_rate: 2e-5
|
105 |
-
# warmup_steps: 100
|
106 |
-
# gradient_checkpointing: true
|
107 |
-
# gradient_accumulation_steps: 4
|
108 |
-
|
109 |
-
# # Hardware optimization
|
110 |
-
# hardware:
|
111 |
-
# mixed_precision: true
|
112 |
-
# cuda_memory_fraction: 0.95
|
113 |
-
# optimize_model_memory: true
|
114 |
-
```
|
|
|
8 |
- merge
|
9 |
|
10 |
---
|
11 |
+
# fmx-3b Augmentive Instruct (preview)
|
12 |
|
13 |
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
14 |
|
|
|
21 |
|
22 |
The following models were included in the merge:
|
23 |
* [unsloth/Qwen2.5-3B-Instruct](https://huggingface.co/unsloth/Qwen2.5-3B-Instruct)
|
24 |
+
* triples
|
25 |
+
* genstruct
|
26 |
+
* kg
|
27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|