Llama3.2 (1B And 3B) Conversational
News
Placeholder
Installation
Unsloth
We now add LoRA adapters so we only need to update 1 to 10% of all parameters!
Unsloth 2024.10.0 patched 28 layers with 28 QKV layers, 28 O layers and 28 MLP layers.
Data Prep
We now use the Llama-3.1 format for conversation style finetunes. We use Maxime Labonne's FineTome-100k dataset in ShareGPT style. But we convert it to HuggingFace's normal multiturn format ("role", "content") instead of ("from", "value")/ Llama-3 renders multi turn conversations like below:
<|begin_of_text|><|start_header_id|>user<|end_header_id|>
Hello!<|eot_id|><|start_header_id|>assistant<|end_header_id|>
Hey there! How are you?<|eot_id|><|start_header_id|>user<|end_header_id|>
I'm great thanks!<|eot_id|>
We use our get_chat_template function to get the correct chat template. We support zephyr, chatml, mistral, llama, alpaca, vicuna, vicuna_old, phi3, llama3 and more.
README.md: 0%| | 0.00/982 [00:00<?, ?B/s]
train-00000-of-00001.parquet: 0%| | 0.00/117M [00:00<?, ?B/s]
Generating train split: 0%| | 0/100000 [00:00<?, ? examples/s]
We now use standardize_sharegpt to convert ShareGPT style datasets into HuggingFace's generic format. This changes the dataset from looking like:
{"from": "system", "value": "You are an assistant"}
{"from": "human", "value": "What is 2+2?"}
{"from": "gpt", "value": "It's 4."}
to
{"role": "system", "content": "You are an assistant"}
{"role": "user", "content": "What is 2+2?"}
{"role": "assistant", "content": "It's 4."}
Standardizing format: 0%| | 0/100000 [00:00<?, ? examples/s]
Map: 0%| | 0/100000 [00:00<?, ? examples/s]
We look at how the conversations are structured for item 5:
[{'content': 'How do astronomers determine the original wavelength of light emitted by a celestial body at rest, which is necessary for measuring its speed using the Doppler effect?',
, 'role': 'user'},
, {'content': 'Astronomers make use of the unique spectral fingerprints of elements found in stars. These elements emit and absorb light at specific, known wavelengths, forming an absorption spectrum. By analyzing the light received from distant stars and comparing it to the laboratory-measured spectra of these elements, astronomers can identify the shifts in these wavelengths due to the Doppler effect. The observed shift tells them the extent to which the light has been redshifted or blueshifted, thereby allowing them to calculate the speed of the star along the line of sight relative to Earth.',
, 'role': 'assistant'}] And we see how the chat template transformed these conversations.
[Notice] Llama 3.1 Instruct's default chat template default adds "Cutting Knowledge Date: December 2023\nToday Date: 26 July 2024", so do not be alarmed!
'<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 July 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nHow do astronomers determine the original wavelength of light emitted by a celestial body at rest, which is necessary for measuring its speed using the Doppler effect?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nAstronomers make use of the unique spectral fingerprints of elements found in stars. These elements emit and absorb light at specific, known wavelengths, forming an absorption spectrum. By analyzing the light received from distant stars and comparing it to the laboratory-measured spectra of these elements, astronomers can identify the shifts in these wavelengths due to the Doppler effect. The observed shift tells them the extent to which the light has been redshifted or blueshifted, thereby allowing them to calculate the speed of the star along the line of sight relative to Earth.<|eot_id|>'
Map (num_proc=2): 0%| | 0/100000 [00:00<?, ? examples/s]
max_steps is given, it will override any value given in num_train_epochs
We also use Unsloth's train_on_completions method to only train on the assistant outputs and ignore the loss on the user's inputs.
Map: 0%| | 0/100000 [00:00<?, ? examples/s]
We verify masking is actually done:
'<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 July 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nHow do astronomers determine the original wavelength of light emitted by a celestial body at rest, which is necessary for measuring its speed using the Doppler effect?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nAstronomers make use of the unique spectral fingerprints of elements found in stars. These elements emit and absorb light at specific, known wavelengths, forming an absorption spectrum. By analyzing the light received from distant stars and comparing it to the laboratory-measured spectra of these elements, astronomers can identify the shifts in these wavelengths due to the Doppler effect. The observed shift tells them the extent to which the light has been redshifted or blueshifted, thereby allowing them to calculate the speed of the star along the line of sight relative to Earth.<|eot_id|>'
' \n\nAstronomers make use of the unique spectral fingerprints of elements found in stars. These elements emit and absorb light at specific, known wavelengths, forming an absorption spectrum. By analyzing the light received from distant stars and comparing it to the laboratory-measured spectra of these elements, astronomers can identify the shifts in these wavelengths due to the Doppler effect. The observed shift tells them the extent to which the light has been redshifted or blueshifted, thereby allowing them to calculate the speed of the star along the line of sight relative to Earth.<|eot_id|>'
We can see the System and Instruction prompts are successfully masked!
GPU = Tesla T4. Max memory = 14.748 GB. 2.635 GB of memory reserved.
==((====))== Unsloth - 2x faster free finetuning | Num GPUs = 1 \\ /| Num examples = 100,000 | Num Epochs = 1 O^O/ \_/ \ Batch size per device = 2 | Gradient Accumulation steps = 4 \ / Total batch size = 8 | Total steps = 60 "-____-" Number of trainable parameters = 24,313,856
446.5262 seconds used for training. 7.44 minutes used for training. Peak reserved memory = 6.531 GB. Peak reserved memory for training = 3.896 GB. Peak reserved memory % of max memory = 44.284 %. Peak reserved memory for training % of max memory = 26.417 %.
Inference
Let's run the model! You can change the instruction and input - leave the output blank!
We use min_p = 0.1 and temperature = 1.5. Read this Tweet for more information on why.
The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
['<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 July 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nContinue the fibonnaci sequence: 1, 1, 2, 3, 5, 8,<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nThe Fibonacci sequence is a series of numbers in which each number is the sum of the two preceding numbers. The sequence is: 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144.<|eot_id|>']
You can also use a TextStreamer for continuous inference - so you can see the generation token by token, instead of waiting the whole time!
The Fibonacci sequence is a series of numbers where each number is the sum of the two preceding numbers. The sequence you provided was: 1, 1, 2, 3, 5, 8, 13 The next number in the sequence would be 21, which is 8 + 13. The sequence continues as: 21, 34, 55, 89, 144, 233.<|eot_id|>
('lora_model/tokenizer_config.json',
, 'lora_model/special_tokens_map.json',
, 'lora_model/tokenizer.json') Now if you want to load the LoRA adapters we just saved for inference, set False to True:
The Eiffel Tower, located in the heart of Paris, stands tall among the city's historic and cultural landmarks. This iron structure, standing at an impressive 324 meters high, offers breathtaking views of the City of Light's iconic landscape. The Eiffel Tower was built for the 1889 World's Fair and has since become a symbol of French engineering and culture.<|eot_id|>
You can also use Hugging Face's AutoModelForPeftCausalLM. Only use this if you do not have unsloth installed. It can be hopelessly slow, since 4bit model downloading is not supported, and Unsloth's inference is 2x faster.
Saving to float16 for VLLM
We also support saving to float16 directly. Select merged_16bit for float16 or merged_4bit for int4. We also allow lora adapters as a fallback. Use push_to_hub_merged to upload to your Hugging Face account! You can go to https://huggingface.co/settings/tokens for your personal tokens.
GGUF / llama.cpp Conversion
To save to GGUF / llama.cpp, we support it natively now! We clone llama.cpp and we default save it to q8_0. We allow all methods like q4_k_m. Use save_pretrained_gguf for local saving and push_to_hub_gguf for uploading to HF.
Some supported quant methods (full list on our Wiki page):
q8_0- Fast conversion. High resource use, but generally acceptable.q4_k_m- Recommended. Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q4_K.q5_k_m- Recommended. Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q5_K.
[NEW] To finetune and auto export to Ollama, try our Ollama notebook