Ke Fang mazzzystar

Anti-hype LLM reading list

Goals: Add links that are reasonable and good explanations of how stuff works. No hype and no vendor content if possible. Practical first-hand accounts of models in prod eagerly sought.

Foundational Concepts

Pre-Transformer Models

声学特征提取

本文介绍如何提取提取声学特征用于Merlin训练。在语音合成中，属于声码器(vocoder)的内容。

Merlin可以使用两种vocoder，STRAIGHT或WORLD。WORLD的目标是提取60-dim MGC, variable-dim BAP (BAP dim: 1 for 16Khz, 5 for 48Khz), 1-dim LF0；STRAIGHT的目标是提取60-dim MGC, 25-dim BAP, 1-dim LF0。

新版本的WORLD_v2还在开发中，目标是提取60-dim MGC, 5-dim BAP, 1-dim LF0(MGC和BAP的维度支持微调)。

由于STRAIGHT的使用有严格的证书限制，本文，主要介绍WORLD。

	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	;;
	;; 作者: 李继刚
	;; 日期: 2025-11-12
	;; 剑名: 圆桌讨论
	;; 剑意: 构建一个以“求真”为目标的结构化对话框架。该框架由一位极具洞察力的主持人
	;; 进行引导，邀请代表不同思想的“典型代表人物”进行一场高强度的、即时响应式的
	;; 深度对话。主持人将在每轮总结时生成视觉化的思考框架(ASCII Chart)，通过
	;; “主动质询” 与“协同共建”，对用户提出的议题进行协同探索，最终生成深刻的、
	;; 结构化的知识网络。

	# train_grpo.py
	#
	# See https://github.com/willccbb/verifiers for ongoing developments
	#
	"""
	citation:

	@misc{brown2025grpodemo,
	title={Granular Format Rewards for Eliciting Mathematical Reasoning Capabilities in Small Language Models},
	author={Brown, William},

	# need install asconnect: pip install asconnect, and openai: pip install openai

	import asconnect
	import os
	import openai

	#api key: https://developer.apple.com/documentation/appstoreconnectapi/creating_api_keys_for_app_store_connect_api
	APPCONN_APIKEY_ID = "xxxxx"
	APPCONN_ISSUER_ID = "xxxxxxx"
	APPCONN_KEY_FILE = "xxxxx.p8"

	# ------------------------------------------------------------------
	# EDIT: I eventually found a faster way to run SD on macOS, via MPSGraph (~0.8s / step on M1 Pro):
	# https://github.com/madebyollin/maple-diffusion
	# The original CoreML-related code & discussion is preserved below :)
	# ------------------------------------------------------------------

	# you too can run stable diffusion on the apple silicon GPU (no ANE sadly)
	#
	# quick test portraits (each took 50 steps x 2s / step ~= 100s on my M1 Pro):
	# * https://i.imgur.com/5ywISvm.png

	import requests
	import os, sys
	import concurrent.futures
	from itertools import repeat

	class XimaScraper:
	def __init__(self, album_no, page_num):
	headers = {
	'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:60.0) Gecko/20100101 Firefox/60.0'
	}

	/ Cosine similarity /
	private func cosineSim(A: [Double], B: [Double]) -> Double {
	return dot(A: A, B: B) / (magnitude(A: A) * magnitude(A: B))
	}

	/ Dot Product /
	private func dot(A: [Double], B: [Double]) -> Double {
	var x: Double = 0
	for i in 0...A.count-1 {
	x += A[i] * B[i]

	%!TEX program = xelatex
	% Font Size:
	% 10pt, 11pt, 12pt
	% Paper Size:
	% a4paper, letterpaper, a5paper, leagalpaper, executivepaper, landscape
	% Font Family:
	% roman, sans
	\documentclass[12pt, a4paper, roman]{moderncv}

	% Style: