Update README.md
#1
by
kasper-piskorski
- opened
README.md
CHANGED
@@ -4,118 +4,19 @@ language:
|
|
4 |
- fr
|
5 |
- es
|
6 |
- pt
|
7 |
-
license: other
|
8 |
-
library_name: transformers
|
9 |
tags:
|
10 |
- falcon3
|
11 |
-
|
|
|
12 |
license_link: https://falconllm.tii.ae/falcon-terms-and-conditions.html
|
13 |
-
model-index:
|
14 |
-
- name: Falcon3-10B-Base
|
15 |
-
results:
|
16 |
-
- task:
|
17 |
-
type: text-generation
|
18 |
-
name: Text Generation
|
19 |
-
dataset:
|
20 |
-
name: IFEval (0-Shot)
|
21 |
-
type: HuggingFaceH4/ifeval
|
22 |
-
args:
|
23 |
-
num_few_shot: 0
|
24 |
-
metrics:
|
25 |
-
- type: inst_level_strict_acc and prompt_level_strict_acc
|
26 |
-
value: 36.48
|
27 |
-
name: strict accuracy
|
28 |
-
source:
|
29 |
-
url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=tiiuae/Falcon3-10B-Base
|
30 |
-
name: Open LLM Leaderboard
|
31 |
-
- task:
|
32 |
-
type: text-generation
|
33 |
-
name: Text Generation
|
34 |
-
dataset:
|
35 |
-
name: BBH (3-Shot)
|
36 |
-
type: BBH
|
37 |
-
args:
|
38 |
-
num_few_shot: 3
|
39 |
-
metrics:
|
40 |
-
- type: acc_norm
|
41 |
-
value: 41.38
|
42 |
-
name: normalized accuracy
|
43 |
-
source:
|
44 |
-
url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=tiiuae/Falcon3-10B-Base
|
45 |
-
name: Open LLM Leaderboard
|
46 |
-
- task:
|
47 |
-
type: text-generation
|
48 |
-
name: Text Generation
|
49 |
-
dataset:
|
50 |
-
name: MATH Lvl 5 (4-Shot)
|
51 |
-
type: hendrycks/competition_math
|
52 |
-
args:
|
53 |
-
num_few_shot: 4
|
54 |
-
metrics:
|
55 |
-
- type: exact_match
|
56 |
-
value: 24.77
|
57 |
-
name: exact match
|
58 |
-
source:
|
59 |
-
url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=tiiuae/Falcon3-10B-Base
|
60 |
-
name: Open LLM Leaderboard
|
61 |
-
- task:
|
62 |
-
type: text-generation
|
63 |
-
name: Text Generation
|
64 |
-
dataset:
|
65 |
-
name: GPQA (0-shot)
|
66 |
-
type: Idavidrein/gpqa
|
67 |
-
args:
|
68 |
-
num_few_shot: 0
|
69 |
-
metrics:
|
70 |
-
- type: acc_norm
|
71 |
-
value: 12.75
|
72 |
-
name: acc_norm
|
73 |
-
source:
|
74 |
-
url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=tiiuae/Falcon3-10B-Base
|
75 |
-
name: Open LLM Leaderboard
|
76 |
-
- task:
|
77 |
-
type: text-generation
|
78 |
-
name: Text Generation
|
79 |
-
dataset:
|
80 |
-
name: MuSR (0-shot)
|
81 |
-
type: TAUR-Lab/MuSR
|
82 |
-
args:
|
83 |
-
num_few_shot: 0
|
84 |
-
metrics:
|
85 |
-
- type: acc_norm
|
86 |
-
value: 14.17
|
87 |
-
name: acc_norm
|
88 |
-
source:
|
89 |
-
url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=tiiuae/Falcon3-10B-Base
|
90 |
-
name: Open LLM Leaderboard
|
91 |
-
- task:
|
92 |
-
type: text-generation
|
93 |
-
name: Text Generation
|
94 |
-
dataset:
|
95 |
-
name: MMLU-PRO (5-shot)
|
96 |
-
type: TIGER-Lab/MMLU-Pro
|
97 |
-
config: main
|
98 |
-
split: test
|
99 |
-
args:
|
100 |
-
num_few_shot: 5
|
101 |
-
metrics:
|
102 |
-
- type: acc
|
103 |
-
value: 36.0
|
104 |
-
name: accuracy
|
105 |
-
source:
|
106 |
-
url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=tiiuae/Falcon3-10B-Base
|
107 |
-
name: Open LLM Leaderboard
|
108 |
---
|
109 |
|
110 |
-
<div align="center">
|
111 |
-
<img src="https://huggingface.co/datasets/tiiuae/documentation-images/resolve/main/general/falco3-logo.png" alt="drawing" width="500"/>
|
112 |
-
</div>
|
113 |
|
114 |
# Falcon3-10B-Base
|
115 |
|
116 |
**Falcon3** family of Open Foundation Models is a set of pretrained and instruct LLMs ranging from 1B to 10B parameters.
|
117 |
|
118 |
-
This repository contains the **Falcon3-10B-Base**. It achieves state-of-the-art results (at
|
119 |
Falcon3-10B-Base supports 4 languages (English, French, Spanish, Portuguese) and a context length of up to 32K.
|
120 |
|
121 |
⚠️ **This is a raw, pretrained model, which should be further finetuned using SFT, RLHF, continued pretraining, etc. for most use cases.**
|
@@ -130,7 +31,7 @@ Falcon3-10B-Base supports 4 languages (English, French, Spanish, Portuguese) and
|
|
130 |
- Uses SwiGLu and RMSNorm
|
131 |
- 32K context length
|
132 |
- 131K vocab size
|
133 |
-
- Depth up-scaled from **Falcon3-7B-Base** with
|
134 |
- Supports EN, FR, ES, PT
|
135 |
- Developed by [Technology Innovation Institute](https://www.tii.ae)
|
136 |
- License: TII Falcon-LLM License 2.0
|
@@ -161,11 +62,8 @@ print(response[0]['generated_text'])
|
|
161 |
<br>
|
162 |
|
163 |
## Benchmarks
|
164 |
-
We report in the following table our internal pipeline benchmarks
|
165 |
-
|
166 |
-
- We report **raw scores**.
|
167 |
-
- We use same batch-size across all models.
|
168 |
-
|
169 |
<table border="1" style="width: 100%; text-align: center; border-collapse: collapse;">
|
170 |
<colgroup>
|
171 |
<col style="width: 10%;">
|
@@ -181,7 +79,7 @@ We report in the following table our internal pipeline benchmarks.
|
|
181 |
<th>Benchmark</th>
|
182 |
<th>Gemma2-9B</th>
|
183 |
<th>Yi1.5-9B</th>
|
184 |
-
<th>Mistral-
|
185 |
<th>Falcon3-10B-Base</th>
|
186 |
</tr>
|
187 |
</thead>
|
@@ -203,7 +101,7 @@ We report in the following table our internal pipeline benchmarks.
|
|
203 |
</tr>
|
204 |
<tr>
|
205 |
<td>IFEval</td>
|
206 |
-
<td>21.
|
207 |
<td>29.1</td>
|
208 |
<td>16.1</td>
|
209 |
<td><b>36.4</b></td>
|
@@ -217,7 +115,7 @@ We report in the following table our internal pipeline benchmarks.
|
|
217 |
<td><b>81.4</b></td>
|
218 |
</tr>
|
219 |
<tr>
|
220 |
-
<td>MATH
|
221 |
<td>10.5</td>
|
222 |
<td>9.2</td>
|
223 |
<td>4.9</td>
|
@@ -240,7 +138,7 @@ We report in the following table our internal pipeline benchmarks.
|
|
240 |
</tr>
|
241 |
<tr>
|
242 |
<td>MUSR (0-shot)</td>
|
243 |
-
<td><b>45.
|
244 |
<td>43.3</td>
|
245 |
<td>39.2</td>
|
246 |
<td>44.2</td>
|
@@ -258,7 +156,7 @@ We report in the following table our internal pipeline benchmarks.
|
|
258 |
<td><b>83.0</b></td>
|
259 |
<td>80.5</td>
|
260 |
<td>82.1</td>
|
261 |
-
<td>79.
|
262 |
</tr>
|
263 |
<tr>
|
264 |
<td>SciQ (0-shot)</td>
|
@@ -284,36 +182,20 @@ We report in the following table our internal pipeline benchmarks.
|
|
284 |
</tbody>
|
285 |
</table>
|
286 |
|
287 |
-
## Useful links
|
288 |
-
- View our [release blogpost](https://huggingface.co/blog/falcon3).
|
289 |
-
- Feel free to join [our discord server](https://discord.gg/fwXpMyGc) if you have any questions or to interact with our researchers and developers.
|
290 |
-
|
291 |
## Technical Report
|
292 |
|
293 |
Coming soon....
|
294 |
|
295 |
## Citation
|
296 |
-
If
|
297 |
-
|
298 |
```
|
299 |
@misc{Falcon3,
|
300 |
-
title = {The Falcon 3
|
301 |
-
|
302 |
-
author = {Falcon-LLM Team},
|
303 |
month = {December},
|
304 |
year = {2024}
|
305 |
}
|
306 |
```
|
307 |
-
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard)
|
308 |
-
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/tiiuae__Falcon3-10B-Base-details)
|
309 |
|
310 |
-
| Metric |Value|
|
311 |
-
|-------------------|----:|
|
312 |
-
|Avg. |27.59|
|
313 |
-
|IFEval (0-Shot) |36.48|
|
314 |
-
|BBH (3-Shot) |41.38|
|
315 |
-
|MATH Lvl 5 (4-Shot)|24.77|
|
316 |
-
|GPQA (0-shot) |12.75|
|
317 |
-
|MuSR (0-shot) |14.17|
|
318 |
-
|MMLU-PRO (5-shot) |36.00|
|
319 |
|
|
|
4 |
- fr
|
5 |
- es
|
6 |
- pt
|
|
|
|
|
7 |
tags:
|
8 |
- falcon3
|
9 |
+
license: other
|
10 |
+
license_name: falcon-llm-license
|
11 |
license_link: https://falconllm.tii.ae/falcon-terms-and-conditions.html
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
---
|
13 |
|
|
|
|
|
|
|
14 |
|
15 |
# Falcon3-10B-Base
|
16 |
|
17 |
**Falcon3** family of Open Foundation Models is a set of pretrained and instruct LLMs ranging from 1B to 10B parameters.
|
18 |
|
19 |
+
This repository contains the **Falcon3-10B-Base**. It achieves state-of-the-art results (at release's time) on reasoning, language understanding, instruction following, code and mathematics tasks.
|
20 |
Falcon3-10B-Base supports 4 languages (English, French, Spanish, Portuguese) and a context length of up to 32K.
|
21 |
|
22 |
⚠️ **This is a raw, pretrained model, which should be further finetuned using SFT, RLHF, continued pretraining, etc. for most use cases.**
|
|
|
31 |
- Uses SwiGLu and RMSNorm
|
32 |
- 32K context length
|
33 |
- 131K vocab size
|
34 |
+
- Depth up-scaled from **Falcon3-7B-Base** with 2 Teratokens of datasets comprising of web, code, STEM, high quality and mutlilingual data using 2048 H100 GPU chips
|
35 |
- Supports EN, FR, ES, PT
|
36 |
- Developed by [Technology Innovation Institute](https://www.tii.ae)
|
37 |
- License: TII Falcon-LLM License 2.0
|
|
|
62 |
<br>
|
63 |
|
64 |
## Benchmarks
|
65 |
+
We report in the following table our internal pipeline benchmarks:
|
66 |
+
|
|
|
|
|
|
|
67 |
<table border="1" style="width: 100%; text-align: center; border-collapse: collapse;">
|
68 |
<colgroup>
|
69 |
<col style="width: 10%;">
|
|
|
79 |
<th>Benchmark</th>
|
80 |
<th>Gemma2-9B</th>
|
81 |
<th>Yi1.5-9B</th>
|
82 |
+
<th>Mistral-NeMo-12B</th>
|
83 |
<th>Falcon3-10B-Base</th>
|
84 |
</tr>
|
85 |
</thead>
|
|
|
101 |
</tr>
|
102 |
<tr>
|
103 |
<td>IFEval</td>
|
104 |
+
<td>21.2</td>
|
105 |
<td>29.1</td>
|
106 |
<td>16.1</td>
|
107 |
<td><b>36.4</b></td>
|
|
|
115 |
<td><b>81.4</b></td>
|
116 |
</tr>
|
117 |
<tr>
|
118 |
+
<td>MATH(4-shot)</td>
|
119 |
<td>10.5</td>
|
120 |
<td>9.2</td>
|
121 |
<td>4.9</td>
|
|
|
138 |
</tr>
|
139 |
<tr>
|
140 |
<td>MUSR (0-shot)</td>
|
141 |
+
<td><b>45.2</b></td>
|
142 |
<td>43.3</td>
|
143 |
<td>39.2</td>
|
144 |
<td>44.2</td>
|
|
|
156 |
<td><b>83.0</b></td>
|
157 |
<td>80.5</td>
|
158 |
<td>82.1</td>
|
159 |
+
<td>79.5</td>
|
160 |
</tr>
|
161 |
<tr>
|
162 |
<td>SciQ (0-shot)</td>
|
|
|
182 |
</tbody>
|
183 |
</table>
|
184 |
|
|
|
|
|
|
|
|
|
185 |
## Technical Report
|
186 |
|
187 |
Coming soon....
|
188 |
|
189 |
## Citation
|
190 |
+
If Falcon3 family were helpful in your work, feel free to give us a cite.
|
191 |
+
|
192 |
```
|
193 |
@misc{Falcon3,
|
194 |
+
title = {The Falcon 3 family of Open Models},
|
195 |
+
author = {TII Team},
|
|
|
196 |
month = {December},
|
197 |
year = {2024}
|
198 |
}
|
199 |
```
|
|
|
|
|
200 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
201 |
|