Text Generation
PEFT
Safetensors
mistral
conversational
Eval Results
dfurman commited on
Commit
8982cd9
1 Parent(s): ce5ddf6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +16 -9
README.md CHANGED
@@ -10,7 +10,14 @@ pipeline_tag: text-generation
10
  base_model: meta-llama/Llama-2-13b-hf
11
  ---
12
 
13
- # llama-2-13b-instruct-v0.1 🦙🐬
 
 
 
 
 
 
 
14
 
15
  This instruction model was built via parameter-efficient QLoRA finetuning of [llama-2-13b](https://huggingface.co/meta-llama/Llama-2-13b-hf) on the first 100k rows of [ehartford/dolphin](https://huggingface.co/datasets/ehartford/dolphin) (an open-source implementation of [Microsoft's Orca](https://www.microsoft.com/en-us/research/publication/orca-progressive-learning-from-complex-explanation-traces-of-gpt-4/)). Finetuning was executed on a single A6000 (48 GB) for roughly 18 hours on the [Lambda Labs](https://cloud.lambdalabs.com/instances) platform.
16
 
@@ -30,9 +37,9 @@ We use state-of-the-art [Language Model Evaluation Harness](https://github.com/E
30
 
31
  * Model license: Llama 2 Community License Agreement
32
  * Basic usage: [notebook](assets/basic_inference_llama_2_13b_dolphin.ipynb)
33
- * Finetuning script: [script](https://github.com/daniel-furman/sft-demos/blob/main/src/sft/one_gpu/llama-2/dolphin/sft-llama-2-13b-instruct-v0.1-peft.py)
34
- * Loss curves: [plot](https://huggingface.co/dfurman/llama-2-13b-instruct-v0.1-peft#finetuning-description)
35
- * Runtime stats: [table](https://huggingface.co/dfurman/llama-2-13b-instruct-v0.1-peft#runtime-tests)
36
 
37
  ## Example prompts and responses
38
 
@@ -41,7 +48,7 @@ Example 1:
41
  **User**:
42
  >You are a helpful assistant. Write me a numbered list of things to do in New York City.\n
43
 
44
- **llama-2-13b-instruct-v0.1-peft**:
45
 
46
  > 1. Visit the iconic Statue of Liberty and Ellis Island.
47
  > 2. Take a stroll through Central Park and enjoy its many attractions.
@@ -61,7 +68,7 @@ Example 2:
61
  **User**:
62
  >You are a helpful assistant. Write a short email inviting my friends to a dinner party on Friday. Respond succinctly.\n"
63
 
64
- **llama-2-13b-instruct-v0.1-peft**:
65
 
66
  > Subject: Friday Dinner Party Invitation
67
  >
@@ -97,7 +104,7 @@ The llama-2-13b models have been modified from a standard transformer in the fol
97
 
98
  This model was trained on a single A6000 (48 GB) for about 18 hours using the [Lambda Labs](https://cloud.lambdalabs.com/instances) platform.
99
 
100
- ![loss curves](https://raw.githubusercontent.com/daniel-furman/sft-demos/main/assets/jul_24_23_1_13_00_log_loss_curves_llama-2-13b-instruct-v0.1.png)
101
 
102
  The above loss curve was generated from the run's private wandb.ai log.
103
 
@@ -138,7 +145,7 @@ notebook_login()
138
  ```
139
 
140
  ```python
141
- peft_model_id = "dfurman/llama-2-13b-instruct-v0.1-peft"
142
  config = PeftConfig.from_pretrained(peft_model_id)
143
 
144
  bnb_config = BitsAndBytesConfig(
@@ -194,7 +201,7 @@ print(tokenizer.decode(output["sequences"][0], skip_special_tokens=True))
194
  | 2.93 | 1x A100 (40 GB SXM) | torch | bfloat16 | 25 |
195
  | 3.24 | 1x A6000 (48 GB) | torch | bfloat16 | 25 |
196
 
197
- The above runtime stats were generated from this [notebook](https://github.com/daniel-furman/sft-demos/blob/main/src/sft/one_gpu/llama-2/dolphin/postprocessing-llama-2-13b-instruct-v0.1-peft.ipynb).
198
 
199
  ## Acknowledgements
200
 
 
10
  base_model: meta-llama/Llama-2-13b-hf
11
  ---
12
 
13
+ <div align="center">
14
+
15
+ <img src="./assets/llama.png" width="150px">
16
+
17
+ </div>
18
+
19
+
20
+ # Llama-2-13B-Instruct-v0.1 🦙🐬
21
 
22
  This instruction model was built via parameter-efficient QLoRA finetuning of [llama-2-13b](https://huggingface.co/meta-llama/Llama-2-13b-hf) on the first 100k rows of [ehartford/dolphin](https://huggingface.co/datasets/ehartford/dolphin) (an open-source implementation of [Microsoft's Orca](https://www.microsoft.com/en-us/research/publication/orca-progressive-learning-from-complex-explanation-traces-of-gpt-4/)). Finetuning was executed on a single A6000 (48 GB) for roughly 18 hours on the [Lambda Labs](https://cloud.lambdalabs.com/instances) platform.
23
 
 
37
 
38
  * Model license: Llama 2 Community License Agreement
39
  * Basic usage: [notebook](assets/basic_inference_llama_2_13b_dolphin.ipynb)
40
+ * Finetuning script: [script](https://github.com/daniel-furman/sft-demos/blob/main/src/sft/one_gpu/llama-2/dolphin/sft-Llama-2-13B-Instruct-v0.1-peft.py)
41
+ * Loss curves: [plot](https://huggingface.co/dfurman/Llama-2-13B-Instruct-v0.1-peft#finetuning-description)
42
+ * Runtime stats: [table](https://huggingface.co/dfurman/Llama-2-13B-Instruct-v0.1-peft#runtime-tests)
43
 
44
  ## Example prompts and responses
45
 
 
48
  **User**:
49
  >You are a helpful assistant. Write me a numbered list of things to do in New York City.\n
50
 
51
+ **Llama-2-13B-Instruct-v0.1-peft**:
52
 
53
  > 1. Visit the iconic Statue of Liberty and Ellis Island.
54
  > 2. Take a stroll through Central Park and enjoy its many attractions.
 
68
  **User**:
69
  >You are a helpful assistant. Write a short email inviting my friends to a dinner party on Friday. Respond succinctly.\n"
70
 
71
+ **Llama-2-13B-Instruct-v0.1-peft**:
72
 
73
  > Subject: Friday Dinner Party Invitation
74
  >
 
104
 
105
  This model was trained on a single A6000 (48 GB) for about 18 hours using the [Lambda Labs](https://cloud.lambdalabs.com/instances) platform.
106
 
107
+ ![loss curves](https://raw.githubusercontent.com/daniel-furman/sft-demos/main/assets/jul_24_23_1_13_00_log_loss_curves_Llama-2-13B-Instruct-v0.1.png)
108
 
109
  The above loss curve was generated from the run's private wandb.ai log.
110
 
 
145
  ```
146
 
147
  ```python
148
+ peft_model_id = "dfurman/Llama-2-13B-Instruct-v0.1-peft"
149
  config = PeftConfig.from_pretrained(peft_model_id)
150
 
151
  bnb_config = BitsAndBytesConfig(
 
201
  | 2.93 | 1x A100 (40 GB SXM) | torch | bfloat16 | 25 |
202
  | 3.24 | 1x A6000 (48 GB) | torch | bfloat16 | 25 |
203
 
204
+ The above runtime stats were generated from this [notebook](https://github.com/daniel-furman/sft-demos/blob/main/src/sft/one_gpu/llama-2/dolphin/postprocessing-Llama-2-13B-Instruct-v0.1-peft.ipynb).
205
 
206
  ## Acknowledgements
207