Update to swift-transformers

[johnmai-dev]

Hi @davidkoski @awni !
Is it possible to upgrade the dependencies?

The latest version of swift-transformers supports chat-template. huggingface/swift-transformers#104

ChatMLX depends on mlx-libraries and the latest version of swift-transformers, but mlx-libraries uses an older swift-transformers version. Looking forward to an upgrade.

mlx-swift-examples/Package.swift

Line 19 in 9266a62

.package(url: "https://github.com/huggingface/swift-transformers", from: "0.1.9"),

After upgrading, you can also modify Libraries/LLM/Models.swift.

mlx-swift-examples/Libraries/LLM/Models.swift

Lines 115 to 237 in 9266a62

	extension ModelConfiguration {
	public static let smolLM_135M_4bit = ModelConfiguration(
	id: "mlx-community/SmolLM-135M-Instruct-4bit",
	defaultPrompt: "Tell me about the history of Spain."
	) {
	prompt in
	"<|im_start|>user\n\(prompt)<|im_end|>\n<|im_start|>assistant\n"
	}
	public static let mistralNeMo4bit = ModelConfiguration(
	id: "mlx-community/Mistral-Nemo-Instruct-2407-4bit",
	defaultPrompt: "Explain quaternions."
	) { prompt in
	"<s>[INST] \(prompt) [/INST] "
	}
	public static let mistral7B4bit = ModelConfiguration(
	id: "mlx-community/Mistral-7B-Instruct-v0.3-4bit",
	defaultPrompt: "Describe the Swift language."
	) { prompt in
	"<s>[INST] \(prompt) [/INST] "
	}
	public static let codeLlama13b4bit = ModelConfiguration(
	id: "mlx-community/CodeLlama-13b-Instruct-hf-4bit-MLX",
	overrideTokenizer: "PreTrainedTokenizer",
	defaultPrompt: "func sortArray(_ array: [Int]) -> String { <FILL_ME> }"
	) { prompt in
	// given the prompt: func sortArray(_ array: [Int]) -> String { <FILL_ME> }
	// the python code produces this (via its custom tokenizer):
	// <PRE> func sortArray(_ array: [Int]) -> String { <SUF> } <MID>
	"<PRE> " + prompt.replacingOccurrences(of: "<FILL_ME>", with: "<SUF>") + " <MID>"
	}
	public static let phi4bit = ModelConfiguration(
	id: "mlx-community/phi-2-hf-4bit-mlx",
	// https://www.promptingguide.ai/models/phi-2
	defaultPrompt: "Why is the sky blue?"
	)
	public static let phi3_5_4bit = ModelConfiguration(
	id: "mlx-community/Phi-3.5-mini-instruct-4bit",
	defaultPrompt: "What is the gravity on Mars and the moon?",
	extraEOSTokens: ["<|end|>"]
	) {
	prompt in
	"<s><|user|>\n\(prompt)<|end|>\n<|assistant|>\n"
	}
	public static let gemma2bQuantized = ModelConfiguration(
	id: "mlx-community/quantized-gemma-2b-it",
	overrideTokenizer: "PreTrainedTokenizer",
	// https://www.promptingguide.ai/models/gemma
	defaultPrompt: "what is the difference between lettuce and cabbage?"
	) { prompt in
	"<start_of_turn>user\n\(prompt)<end_of_turn>\n<start_of_turn>model\n"
	}
	public static let gemma_2_9b_it_4bit = ModelConfiguration(
	id: "mlx-community/gemma-2-9b-it-4bit",
	overrideTokenizer: "PreTrainedTokenizer",
	// https://www.promptingguide.ai/models/gemma
	defaultPrompt: "What is the difference between lettuce and cabbage?"
	) { prompt in
	"<start_of_turn>user\n\(prompt)<end_of_turn>\n<start_of_turn>model\n"
	}
	public static let gemma_2_2b_it_4bit = ModelConfiguration(
	id: "mlx-community/gemma-2-2b-it-4bit",
	overrideTokenizer: "PreTrainedTokenizer",
	// https://www.promptingguide.ai/models/gemma
	defaultPrompt: "What is the difference between lettuce and cabbage?"
	) { prompt in
	"<start_of_turn>user \(prompt)<end_of_turn><start_of_turn>model"
	}
	public static let qwen205b4bit = ModelConfiguration(
	id: "mlx-community/Qwen1.5-0.5B-Chat-4bit",
	overrideTokenizer: "PreTrainedTokenizer",
	defaultPrompt: "why is the sky blue?"
	) { prompt in
	"<|im_start|>system\nYou are a helpful assistant<|im_end|>\n<|im_start|>user\n\(prompt)<|im_end|>\n<|im_start|>assistant"
	}
	public static let openelm270m4bit = ModelConfiguration(
	id: "mlx-community/OpenELM-270M-Instruct",
	// https://huggingface.co/apple/OpenELM
	defaultPrompt: "Once upon a time there was"
	) { prompt in
	"\(prompt)"
	}
	public static let llama3_1_8B_4bit = ModelConfiguration(
	id: "mlx-community/Meta-Llama-3.1-8B-Instruct-4bit",
	defaultPrompt: "What is the difference between a fruit and a vegetable?"
	) {
	prompt in
	"<|begin_of_text|><|start_header_id|>system<|end_header_id|>\nYou are a helpful assistant<|eot_id|>\n<|start_header_id|>user<|end_header_id|>\n\(prompt)<|eot_id|>\n<|start_header_id|>assistant<|end_header_id|>"
	}
	public static let llama3_8B_4bit = ModelConfiguration(
	id: "mlx-community/Meta-Llama-3-8B-Instruct-4bit",
	defaultPrompt: "What is the difference between a fruit and a vegetable?"
	) {
	prompt in
	"<|begin_of_text|><|start_header_id|>system<|end_header_id|>\nYou are a helpful assistant<|eot_id|>\n<|start_header_id|>user<|end_header_id|>\n\(prompt)<|eot_id|>\n<|start_header_id|>assistant<|end_header_id|>"
	}
	private enum BootstrapState: Sendable {
	case idle
	case bootstrapping
	case bootstrapped
	}

[davidkoski]

Yes of course! If you want to submit a PR that is the fastest way, otherwise I will try to get to this in the coming week.

[johnmai-dev]

Okay, thank you very much @davidkoski!
I can only modify mlx-swift-examples/Package.swift on my end.

As for modifying mlx-swift-examples/Libraries/LLM/Models.swift, I might not have time for that at the moment.

Hi @pcuenca, could you please tag a new release for https://github.com/huggingface/swift-transformers?

[davidkoski]

I think we are all set here