johannaSommer commited on
Commit
f708ae1
1 Parent(s): 201dc1f

1d7c6ad0957b66aee9cb6ac3b88cc411b600a3e922aeea0d3b151b95a897c433

Browse files
Files changed (4) hide show
  1. README.md +9 -9
  2. base_results.json +11 -0
  3. plots.png +0 -0
  4. smashed_results.json +11 -0
README.md CHANGED
@@ -1,6 +1,6 @@
1
  ---
2
  thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
3
- base_model: TinyLlama/TinyLlama_v1.1_math_code
4
  metrics:
5
  - memory_disk
6
  - memory_inference
@@ -38,11 +38,11 @@ tags:
38
  ![image info](./plots.png)
39
 
40
  **Frequently Asked Questions**
41
- - ***How does the compression work?*** The model is compressed with awq.
42
  - ***How does the model quality change?*** The quality of the model output might vary compared to the base model.
43
  - ***How is the model efficiency evaluated?*** These results were obtained on HARDWARE_NAME with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
44
  - ***What is the model format?*** We use safetensors.
45
- - ***What calibration data has been used?*** If needed by the compression method, we used WikiText as the calibration data.
46
  - ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
47
  - ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
48
  - ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
@@ -52,18 +52,18 @@ tags:
52
 
53
  You can run the smashed model with these steps:
54
 
55
- 0. Check requirements from the original repo TinyLlama/TinyLlama_v1.1_math_code installed. In particular, check python, cuda, and transformers versions.
56
  1. Make sure that you have installed quantization related packages.
57
  ```bash
58
- pip install autoawq
59
  ```
60
  2. Load & run the model.
61
  ```python
62
  from transformers import AutoModelForCausalLM, AutoTokenizer
63
- from awq import AutoAWQForCausalLM
64
 
65
- model = AutoAWQForCausalLM.from_quantized("PrunaAI/TinyLlama-TinyLlama_v1.1_math_code-AWQ-4bit-smashed", trust_remote_code=True, device_map='auto')
66
- tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama_v1.1_math_code")
67
 
68
  input_ids = tokenizer("What is the color of prunes?,", return_tensors='pt').to(model.device)["input_ids"]
69
 
@@ -77,7 +77,7 @@ The configuration info are in `smash_config.json`.
77
 
78
  ## Credits & License
79
 
80
- The license of the smashed model follows the license of the original model. Please check the license of the original model TinyLlama/TinyLlama_v1.1_math_code before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
81
 
82
  ## Want to compress other models?
83
 
 
1
  ---
2
  thumbnail: "https://assets-global.website-files.com/646b351987a8d8ce158d1940/64ec9e96b4334c0e1ac41504_Logo%20with%20white%20text.svg"
3
+ base_model: ORIGINAL_REPO_NAME
4
  metrics:
5
  - memory_disk
6
  - memory_inference
 
38
  ![image info](./plots.png)
39
 
40
  **Frequently Asked Questions**
41
+ - ***How does the compression work?*** The model is compressed with COMPRESSION_METHODS.
42
  - ***How does the model quality change?*** The quality of the model output might vary compared to the base model.
43
  - ***How is the model efficiency evaluated?*** These results were obtained on HARDWARE_NAME with configuration described in `model/smash_config.json` and are obtained after a hardware warmup. The smashed model is directly compared to the original base model. Efficiency results may vary in other settings (e.g. other hardware, image size, batch size, ...). We recommend to directly run them in the use-case conditions to know if the smashed model can benefit you.
44
  - ***What is the model format?*** We use safetensors.
45
+ - ***What calibration data has been used?*** If needed by the compression method, we used CALIBRATION_DATASET as the calibration data.
46
  - ***What is the naming convention for Pruna Huggingface models?*** We take the original model name and append "turbo", "tiny", or "green" if the smashed model has a measured inference speed, inference memory, or inference energy consumption which is less than 90% of the original base model.
47
  - ***How to compress my own models?*** You can request premium access to more compression methods and tech support for your specific use-cases [here](https://z0halsaff74.typeform.com/pruna-access?typeform-source=www.pruna.ai).
48
  - ***What are "first" metrics?*** Results mentioning "first" are obtained after the first run of the model. The first run might take more memory or be slower than the subsequent runs due cuda overheads.
 
52
 
53
  You can run the smashed model with these steps:
54
 
55
+ 0. Check requirements from the original repo ORIGINAL_REPO_NAME installed. In particular, check python, cuda, and transformers versions.
56
  1. Make sure that you have installed quantization related packages.
57
  ```bash
58
+ REQUIREMENTS_INSTRUCTIONS
59
  ```
60
  2. Load & run the model.
61
  ```python
62
  from transformers import AutoModelForCausalLM, AutoTokenizer
63
+ IMPORTS
64
 
65
+ MODEL_LOAD
66
+ tokenizer = AutoTokenizer.from_pretrained("ORIGINAL_REPO_NAME")
67
 
68
  input_ids = tokenizer("What is the color of prunes?,", return_tensors='pt').to(model.device)["input_ids"]
69
 
 
77
 
78
  ## Credits & License
79
 
80
+ The license of the smashed model follows the license of the original model. Please check the license of the original model ORIGINAL_REPO_NAME before using this model which provided the base model. The license of the `pruna-engine` is [here](https://pypi.org/project/pruna-engine/) on Pypi.
81
 
82
  ## Want to compress other models?
83
 
base_results.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "current_gpu_type": "NVIDIA A10G",
3
+ "current_gpu_total_memory": 22515.75,
4
+ "memory_inference_first": 4198.0,
5
+ "memory_inference": 4198.0,
6
+ "memory_disk_first": 4196.0,
7
+ "memory_disk": 1258.0,
8
+ "inference_latency_async": 21.704881191253662,
9
+ "inference_CO2_emissions": 3.445414542924158e-05,
10
+ "inference_energy_consumption": 0.0001193449333505799
11
+ }
plots.png ADDED
smashed_results.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "current_gpu_type": "NVIDIA A10G",
3
+ "current_gpu_total_memory": 22515.75,
4
+ "memory_inference_first": 778.0,
5
+ "memory_inference": 762.0,
6
+ "memory_disk_first": 782.0,
7
+ "memory_disk": 762.0,
8
+ "inference_latency_async": 34.14717197418213,
9
+ "inference_CO2_emissions": 3.281787101072708e-05,
10
+ "inference_energy_consumption": 0.00011251503634180266
11
+ }