Skip to content

sampling v2 update #37

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 19, 2024
Merged

sampling v2 update #37

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 14 additions & 18 deletions Sources/LLM/LLM.swift
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@ open class LLM: ObservableObject {
}
}

public var seed: UInt32
public var topK: Int32
public var topP: Float
public var temp: Float
Expand Down Expand Up @@ -74,13 +75,13 @@ open class LLM: ObservableObject {
#endif
let model = llama_load_model_from_file(self.path, modelParams)!
params = llama_context_default_params()
let processorCount = UInt32(ProcessInfo().processorCount)
let processorCount = Int32(ProcessInfo().processorCount)
self.maxTokenCount = Int(min(maxTokenCount, llama_n_ctx_train(model)))
params.seed = seed
params.n_ctx = UInt32(self.maxTokenCount)
params.n_batch = params.n_ctx
params.n_threads = processorCount
params.n_threads_batch = processorCount
self.seed = seed
self.topK = topK
self.topP = topP
self.temp = temp
Expand Down Expand Up @@ -186,22 +187,17 @@ open class LLM: ObservableObject {
@InferenceActor
private func predictNextToken() async -> Token {
guard shouldContinuePredicting else { return model.endToken }
let logits = llama_get_logits_ith(context.pointer, batch.n_tokens - 1)!
var candidates = (0..<totalTokenCount).map { token in
llama_token_data(id: Int32(token), logit: logits[token], p: 0.0)
}
var token: llama_token!
candidates.withUnsafeMutableBufferPointer { pointer in
var candidates = llama_token_data_array(
data: pointer.baseAddress,
size: totalTokenCount,
sorted: false
)
llama_sample_top_k(context.pointer, &candidates, topK, 1)
llama_sample_top_p(context.pointer, &candidates, topP, 1)
llama_sample_temp(context.pointer, &candidates, temp)
token = llama_sample_token(context.pointer, &candidates)
}
let samplerParams = llama_sampler_chain_default_params()
let sampler = llama_sampler_chain_init(samplerParams)

llama_sampler_chain_add(sampler, llama_sampler_init_top_k(topK))
llama_sampler_chain_add(sampler, llama_sampler_init_top_p(topP, 1))
llama_sampler_chain_add(sampler, llama_sampler_init_temp(temp))
llama_sampler_chain_add(sampler, llama_sampler_init_dist(seed))

let i = batch.n_tokens - 1
let token = llama_sampler_sample(sampler, context.pointer, i)

batch.clear()
batch.add(token, currentCount, [0], true)
context.decode(batch)
Expand Down