tensorplex-labs
commited on
Commit
•
70a4ed5
1
Parent(s):
61a5408
Update README.md
Browse files
README.md
CHANGED
@@ -11,7 +11,6 @@ tags:
|
|
11 |
- base-model
|
12 |
- bittensor
|
13 |
- decentralized AI
|
14 |
-
- Web3
|
15 |
datasets:
|
16 |
- tiiuae/falcon-refinedweb
|
17 |
---
|
@@ -44,18 +43,17 @@ Since the parameter limit was upgraded to 7 billion on April 19, 2024, Tensorple
|
|
44 |
- **Architecture**: Adopted Llama-style architecture with 6.9 billion parameters
|
45 |
- **Training Data**: Trained on the tiiuae/falcon-refinedweb dataset
|
46 |
- **Training Objective**: Causal Language Modeling (next token prediction)
|
|
|
47 |
|
48 |
Sumo-Qyuu-7B-v0.1 features a larger vocabulary size (100k), compatible with the GPT-4 tokenizer, ensuring its versatility across various natural language processing tasks.
|
49 |
|
|
|
|
|
50 |
### Model Sources
|
51 |
|
52 |
- **Bittensor Subnet9 Leaderboard:** [https://huggingface.co/spaces/RaoFoundation/pretraining-leaderboard](https://huggingface.co/spaces/RaoFoundation/pretraining-leaderboard)
|
53 |
- **Bittensor Subnet9 Repository:** [https://github.com/RaoFoundation/pretraining/tree/main](https://github.com/RaoFoundation/pretraining/tree/main)
|
54 |
|
55 |
-
## Usage
|
56 |
-
|
57 |
-
⛔ **This is a pretrained base model, which hasn't been aligned yet. Use with caution or finetune further on downstream tasks before deployment.**
|
58 |
-
|
59 |
## How to Get Started with the Model
|
60 |
|
61 |
Use the code below to get started with the model.
|
@@ -93,26 +91,23 @@ for seq in sequences:
|
|
93 |
|
94 |
### Training Data
|
95 |
|
96 |
-
This model has been trained with [tiiuae/falcon-refinedweb](https://huggingface.co/datasets/tiiuae/falcon-refinedweb) dataset continuously.
|
97 |
|
98 |
## Evaluation
|
99 |
|
100 |
Sumo-Qyuu-7B-v0.1 has outperformed notable models such as TII Falcon 7B, Meta's Llama-2-7b and Llama-1-7b in zero-shot performance,
|
101 |
establishing itself as the leading model in aggregate across various evaluation tasks.
|
102 |
-
Such benchmarks include ARC Challenge, GSM8K, HellaSwag, MMLU, TruthfulQA
|
103 |
|
104 |
|
105 |
-
|
|
106 |
-
|
107 |
-
|
|
108 |
-
|
|
109 |
-
|
|
110 |
-
|
|
111 |
-
|
|
112 |
-
| truthfulqa_mc2 (acc, 0-shot) | 37.29 | 39.00 | 34.01 | 34.27 |
|
113 |
-
| winogrande (acc, 0-shot) | 70.88 | 68.67 | 70.17 | 67.17 |
|
114 |
|
115 |
-
[LM Evaluation Harness Repository](https://github.com/EleutherAI/lm-evaluation-harness)
|
116 |
|
117 |
## Future Plans
|
118 |
|
|
|
11 |
- base-model
|
12 |
- bittensor
|
13 |
- decentralized AI
|
|
|
14 |
datasets:
|
15 |
- tiiuae/falcon-refinedweb
|
16 |
---
|
|
|
43 |
- **Architecture**: Adopted Llama-style architecture with 6.9 billion parameters
|
44 |
- **Training Data**: Trained on the tiiuae/falcon-refinedweb dataset
|
45 |
- **Training Objective**: Causal Language Modeling (next token prediction)
|
46 |
+
- **Original Model Repo**: [tensorplex-labs/pretraining-sn9-7B-1](https://huggingface.co/tensorplex-labs/pretraining-sn9-7B-1)
|
47 |
|
48 |
Sumo-Qyuu-7B-v0.1 features a larger vocabulary size (100k), compatible with the GPT-4 tokenizer, ensuring its versatility across various natural language processing tasks.
|
49 |
|
50 |
+
⛔ **This is a pretrained base model, which hasn't been aligned yet. Use with caution or finetune further on downstream tasks before deployment.**
|
51 |
+
|
52 |
### Model Sources
|
53 |
|
54 |
- **Bittensor Subnet9 Leaderboard:** [https://huggingface.co/spaces/RaoFoundation/pretraining-leaderboard](https://huggingface.co/spaces/RaoFoundation/pretraining-leaderboard)
|
55 |
- **Bittensor Subnet9 Repository:** [https://github.com/RaoFoundation/pretraining/tree/main](https://github.com/RaoFoundation/pretraining/tree/main)
|
56 |
|
|
|
|
|
|
|
|
|
57 |
## How to Get Started with the Model
|
58 |
|
59 |
Use the code below to get started with the model.
|
|
|
91 |
|
92 |
### Training Data
|
93 |
|
94 |
+
This model has been trained with [tiiuae/falcon-refinedweb](https://huggingface.co/datasets/tiiuae/falcon-refinedweb) dataset, and still ongoing continuously.
|
95 |
|
96 |
## Evaluation
|
97 |
|
98 |
Sumo-Qyuu-7B-v0.1 has outperformed notable models such as TII Falcon 7B, Meta's Llama-2-7b and Llama-1-7b in zero-shot performance,
|
99 |
establishing itself as the leading model in aggregate across various evaluation tasks.
|
100 |
+
Such benchmarks include ARC Challenge, GSM8K, HellaSwag, MMLU, TruthfulQA, and Winogrande.
|
101 |
|
102 |
|
103 |
+
| | avg | arc_challenge | gsm8k | hellaswag | mmlu | truthfulqa_mc2 | winogrande |
|
104 |
+
|:--------------------------------------|-----------:|----------------:|--------:|------------:|-------:|-----------------:|-------------:|
|
105 |
+
| meta-llama/Meta-Llama-3-8B | 0.6009 | 0.5333 | 0.4913 | 0.7906 | 0.621 | 0.4392 | 0.7301 |
|
106 |
+
| **tensorplex-labs/Sumo-Qyuu-7B-v0.1** | **0.4769** | 0.4753 | 0.1031 | 0.7666 | 0.4426 | 0.3723 | 0.7017 |
|
107 |
+
| meta-llama/Llama-2-7b-hf | 0.473 | 0.4625 | 0.1213 | 0.7597 | 0.4123 | 0.3896 | 0.693 |
|
108 |
+
| huggyllama/llama-7b | 0.4386 | 0.4471 | 0.0849 | 0.7621 | 0.2973 | 0.3408 | 0.6993 |
|
109 |
+
| tiiuae/falcon-7b | 0.4189 | 0.4343 | 0.0432 | 0.7636 | 0.2582 | 0.3428 | 0.6717 |
|
|
|
|
|
110 |
|
|
|
111 |
|
112 |
## Future Plans
|
113 |
|