使用 LoRA 对 Gemma 2 进行微调，以纳入 Rust 官方文档

# Install Keras 3 last. See https://keras.io/getting_started/ for more details.!pip install -q -U keras-nlp!pip install -q -U keras>=3

import os
os.environ["KERAS_BACKEND"] = "jax"  # Or "torch" or "tensorflow".# Avoid memory fragmentation on JAX backend.os.environ["XLA_PYTHON_CLIENT_MEM_FRACTION"]="1.00"

import kerasimport keras_nlp

template = "Instruction:\n{question}\n\nResponse:\n{answer}"

gemma_lm = keras_nlp.models.GemmaCausalLM.from_preset("gemma2_instruct_2b_en")gemma_lm.summary()

Preprocessor: "gemma_causal_lm_preprocessor"

Model: "gemma_causal_lm"

Total params: 2,614,341,888 (9.74 GB)

Trainable params: 2,614,341,888 (9.74 GB)

Non-trainable params: 0 (0.00 B)

prompt = template.format(    question="How can I overload the `+` operator for arithmetic addition in Rust?",    answer="",)print(gemma_lm.generate(prompt, max_length=256))

Instruction:How can I overload the `+` operator for arithmetic addition in Rust?Response:```ruststruct Point {    x: f64,    y: f64,}impl Point {    fn new(x: f64, y: f64) -> Self {        Point { x, y }    }    fn add(self, other: Point) -> Point {        Point {            x: self.x + other.x,            y: self.y + other.y,        }    }}fn main() {    let p1 = Point::new(1.0, 2.0);    let p2 = Point::new(3.0, 4.0);    let result = p1 + p2;    println!("Result: ({}, {})", result.x, result.y);}```**Explanation:**1. **Struct Definition:** We define a `Point` struct to represent points in 2D space.2. **`add` Method:** We implement the `+` operator for the `Point`

import jsondata = []with open('/kaggle/input/rust-official-book/dataset.jsonl', encoding='utf-8') as file:    for line in file:        features = json.loads(line)        # Format the entire example as a single string.        data.append(template.format(**features))
# Only use 1000 training examples, to keep it fast.# data = data[:100]

# Enable LoRA for the model and set the LoRA rank to 4.gemma_lm.backbone.enable_lora(rank=4)gemma_lm.summary()

Preprocessor: "gemma_causal_lm_preprocessor"

Model: "gemma_causal_lm"

Total params: 2,617,270,528 (9.75 GB)

Trainable params: 2,928,640 (11.17 MB)

Non-trainable params: 2,614,341,888 (9.74 GB)

# Limit the input sequence length to 512 (to control memory usage).gemma_lm.preprocessor.sequence_length = 512# Use AdamW (a common optimizer for transformer models).optimizer = keras.optimizers.AdamW(    learning_rate=5e-5,    weight_decay=0.01,)# Exclude layernorm and bias terms from decay.optimizer.exclude_from_weight_decay(var_names=["bias", "scale"])
gemma_lm.compile(    loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),    optimizer=optimizer,    weighted_metrics=[keras.metrics.SparseCategoricalAccuracy()],)gemma_lm.fit(data, epochs=1, batch_size=1)

prompt = template.format(    question="How can I overload the `+` operator for arithmetic addition in Rust?",    answer="",)print(gemma_lm.generate(prompt, max_length=256))