CoolFace
Modelpublic

keras/llama3_8b_en_int8

sourceHugging Facellama3updated 1y agoView on Hugging Face
0likes10downloads
Model Card

Model Overview

Llama 3 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.

Weights are released under the Llama 3 Community License. Keras model code is released under the Apache 2 License.

Links

Installation

Keras and KerasHub can be installed with:

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 nameParametersDescription
llama3_8b_en 8.03B8 billion parameter, 32-layer, base LLaMA 3 model.
llama3_8b_en_int8 8.03B8 billion parameter, 32-layer, base LLaMA 3 model with activation and weights quantized to int8.
llama3_instruct_8b_en 8.03B8 billion parameter, 32-layer, instruction tuned LLaMA 3 model.
llama3_instruct_8b_en_int8 8.03B8 billion parameter, 32-layer, instruction tuned LLaMA 3 model with activation and weights quantized to int8.
llama3.1_8b8.03B8 billion parameter, 32-layer, based LLaMA 3.1 model.
llama3.1_guard_8b8.03B8 billion parameter, 32-layer, LLaMA 3.1 fine-tuned for consent safety classification.
llama3.1_instruct_8b8.03B8 billion parameter, 32-layer, instruction tuned LLaMA 3.1.
llama3.2_1b1.5B1 billion parameter, 16-layer, based LLaMA 3.2 model.
llama3.2_3b3.6B3 billion parameter, 26-layer, based LLaMA 3.2 model.
llama3.2_guard_1b1.5B1 billion parameter, 16-layer, based LLaMA 3.2 model fine-tuned for consent safety classification.
llama3.2_instruct_1b1.5B1 billion parameter, 16-layer, instruction tuned LLaMA 3.2.
llama3.2_instruct_3b3.6B3 billion parameter, 28-layer, instruction tuned LLaMA 3.2.

Prompts

Llama-3 "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:

python
prompt = """<|start_header_id|>system<|end_header_id|>

You are a helpful AI assistant for travel tips and recommendations<|eot_id|><|start_header_id|>user<|end_header_id|>

What can you help me with?<|eot_id|><|start_header_id|>assistant<|end_header_id|>
"""

For more details, please refer to this link: Llama 3 Model Card & Prompt Formats.

Base models (without instruct in the name) have no specific prompting structure, and should usually be fine-tuned for a specific task.

Example Usage

python
import keras
import keras_hub
import numpy as np

Use generate() to do text generation.

python
llama_lm = keras_hub.models.Llama3CausalLM.from_preset("llama3_8b_en_int8")
llama_lm.generate("What is Keras?", max_length=500)

# Generate with batched prompts.
llama_lm.generate(["What is Keras?", "Give me your best brownie recipe."], max_length=500)

Compile the generate() function with a custom sampler.

python
llama_lm = keras_hub.models.Llama3CausalLM.from_preset("llama3_8b_en_int8")
llama_lm.compile(sampler="greedy")
llama_lm.generate("I want to say", max_length=30)

llama_lm.compile(sampler=keras_hub.samplers.BeamSampler(num_beams=2))
llama_lm.generate("I want to say", max_length=30)

Use generate() without preprocessing.

python
prompt = {
    "token_ids": np.array([[306, 864, 304, 1827, 0, 0, 0, 0, 0, 0]] * 2),
    # Use `"padding_mask"` to indicate values that should not be overridden.
    "padding_mask": np.array([[1, 1, 1, 1, 0, 0, 0, 0, 0, 0]] * 2),
}

llama_lm = keras_hub.models.Llama3CausalLM.from_preset(
    "llama3_8b_en_int8",
    preprocessor=None,
    dtype="bfloat16"
)
llama_lm.generate(prompt)

Call fit() on a single batch.

python
features = ["The quick brown fox jumped.", "I forgot my homework."]
llama_lm = keras_hub.models.Llama3CausalLM.from_preset("llama3_8b_en_int8")
llama_lm.fit(x=features, batch_size=2)

Call fit() without preprocessing.

python
x = {
    "token_ids": np.array([[450, 4996, 17354, 1701, 29916, 12500, 287, 29889, 0, 0]] * 2),
    "padding_mask": np.array([[1, 1, 1, 1, 1, 1, 1, 1, 0, 0]] * 2),
}
y = np.array([[4996, 17354, 1701, 29916, 12500, 287, 29889, 0, 0, 0]] * 2)
sw = np.array([[1, 1, 1, 1, 1, 1, 1, 0, 0, 0]] * 2)

llama_lm = keras_hub.models.Llama3CausalLM.from_preset(
    "llama3_8b_en_int8",
    preprocessor=None,
    dtype="bfloat16"
)
llama_lm.fit(x=x, y=y, sample_weight=sw, batch_size=2)

Example Usage with Hugging Face URI

python
import keras
import keras_hub
import numpy as np

Use generate() to do text generation.

python
llama_lm = keras_hub.models.Llama3CausalLM.from_preset("hf://keras/llama3_8b_en_int8")
llama_lm.generate("What is Keras?", max_length=500)

# Generate with batched prompts.
llama_lm.generate(["What is Keras?", "Give me your best brownie recipe."], max_length=500)

Compile the generate() function with a custom sampler.

python
llama_lm = keras_hub.models.Llama3CausalLM.from_preset("hf://keras/llama3_8b_en_int8")
llama_lm.compile(sampler="greedy")
llama_lm.generate("I want to say", max_length=30)

llama_lm.compile(sampler=keras_hub.samplers.BeamSampler(num_beams=2))
llama_lm.generate("I want to say", max_length=30)

Use generate() without preprocessing.

python
prompt = {
    "token_ids": np.array([[306, 864, 304, 1827, 0, 0, 0, 0, 0, 0]] * 2),
    # Use `"padding_mask"` to indicate values that should not be overridden.
    "padding_mask": np.array([[1, 1, 1, 1, 0, 0, 0, 0, 0, 0]] * 2),
}

llama_lm = keras_hub.models.Llama3CausalLM.from_preset(
    "hf://keras/llama3_8b_en_int8",
    preprocessor=None,
    dtype="bfloat16"
)
llama_lm.generate(prompt)

Call fit() on a single batch.

python
features = ["The quick brown fox jumped.", "I forgot my homework."]
llama_lm = keras_hub.models.Llama3CausalLM.from_preset("hf://keras/llama3_8b_en_int8")
llama_lm.fit(x=features, batch_size=2)

Call fit() without preprocessing.

python
x = {
    "token_ids": np.array([[450, 4996, 17354, 1701, 29916, 12500, 287, 29889, 0, 0]] * 2),
    "padding_mask": np.array([[1, 1, 1, 1, 1, 1, 1, 1, 0, 0]] * 2),
}
y = np.array([[4996, 17354, 1701, 29916, 12500, 287, 29889, 0, 0, 0]] * 2)
sw = np.array([[1, 1, 1, 1, 1, 1, 1, 0, 0, 0]] * 2)

llama_lm = keras_hub.models.Llama3CausalLM.from_preset(
    "hf://keras/llama3_8b_en_int8",
    preprocessor=None,
    dtype="bfloat16"
)
llama_lm.fit(x=x, y=y, sample_weight=sw, batch_size=2)