Llama 2 is a set of large language models published by Meta. Both pretrained and instruction tuned models are available, and range in size from 7 billion to 70 billion parameters. See the model card below for benchmarks, data sources, and intended use cases.
pip install -U -q keras-hub
pip install -U -q keras
Jax, TensorFlow, and Torch come preinstalled in Kaggle Notebooks. For instructions on installing them in another environment see the Keras Getting Started page.
Presets
The following model checkpoints are provided by the Keras team. Full code examples for each are available below.
Preset name
Parameters
Description
llama2_7b_en
6.74B
7 billion parameter, 32-layer, base LLaMA 2 model.
llama2_7b_en_int8
6.74B
7 billion parameter, 32-layer, base LLaMA 2 model with activation and weights quantized to int8.
7 billion parameter, 32-layer, instruction tuned LLaMA 2 model with activation and weights quantized to int8.
Prompts
Llama-2 "instruct" models are instruction tuned on turn by turn conversations and should be prompted with examples that precisely match the training data. Specifically, you must alternate user and assistant turns that begin and end with special tokens. New lines do matter. See the following for an example:
prompt = """<s>[INST] <
Example Usage
python
1import keras
2import keras_hub
3import numpy as np
Use generate() to do text generation.
python
1llama_lm = keras_hub.models.LlamaCausalLM.from_preset("llama2_instruct_7b_en")2llama_lm.generate("What is Keras?", max_length=500)34# Generate with batched prompts.5llama_lm.generate(["What is Keras?","Give me your best brownie recipe."], max_length=500)
Compile the generate() function with a custom sampler.
python
1llama_lm = keras_hub.models.LlamaCausalLM.from_preset("llama2_instruct_7b_en")2llama_lm.compile(sampler="greedy")3llama_lm.generate("I want to say", max_length=30)45llama_lm.compile(sampler=keras_hub.samplers.BeamSampler(num_beams=2))6llama_lm.generate("I want to say", max_length=30)
Use generate() without preprocessing.
python
1prompt ={2# `1` maps to the start token followed by "I want to say".3"token_ids": np.array([[1,306,864,304,1827,0,0,0,0,0]]*2),4# Use `"padding_mask"` to indicate values that should not be overridden.5"padding_mask": np.array([[1,1,1,1,1,0,0,0,0,0]]*2),6}78llama_lm = keras_hub.models.LlamaCausalLM.from_preset(9"llama2_instruct_7b_en",10 preprocessor=None,11 dtype="bfloat16"12)13llama_lm.generate(prompt)
Call fit() on a single batch.
python
1features =["The quick brown fox jumped.","I forgot my homework."]2llama_lm = keras_hub.models.LlamaCausalLM.from_preset("llama2_instruct_7b_en")3llama_lm.fit(x=features, batch_size=2)
1import keras
2import keras_hub
3import numpy as np
Use generate() to do text generation.
python
1llama_lm = keras_hub.models.LlamaCausalLM.from_preset("hf://keras/llama2_instruct_7b_en")2llama_lm.generate("What is Keras?", max_length=500)34# Generate with batched prompts.5llama_lm.generate(["What is Keras?","Give me your best brownie recipe."], max_length=500)
Compile the generate() function with a custom sampler.
python
1llama_lm = keras_hub.models.LlamaCausalLM.from_preset("hf://keras/llama2_instruct_7b_en")2llama_lm.compile(sampler="greedy")3llama_lm.generate("I want to say", max_length=30)45llama_lm.compile(sampler=keras_hub.samplers.BeamSampler(num_beams=2))6llama_lm.generate("I want to say", max_length=30)
Use generate() without preprocessing.
python
1prompt ={2# `1` maps to the start token followed by "I want to say".3"token_ids": np.array([[1,306,864,304,1827,0,0,0,0,0]]*2),4# Use `"padding_mask"` to indicate values that should not be overridden.5"padding_mask": np.array([[1,1,1,1,1,0,0,0,0,0]]*2),6}78llama_lm = keras_hub.models.LlamaCausalLM.from_preset(9"hf://keras/llama2_instruct_7b_en",10 preprocessor=None,11 dtype="bfloat16"12)13llama_lm.generate(prompt)
Call fit() on a single batch.
python
1features =["The quick brown fox jumped.","I forgot my homework."]2llama_lm = keras_hub.models.LlamaCausalLM.from_preset("hf://keras/llama2_instruct_7b_en")3llama_lm.fit(x=features, batch_size=2)
