File size: 12,639 Bytes
3c9a12b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 | calibration_set:
_templates:
programming_languages: &programming_languages "Solve the following problem using {{ ['Zephyr', 'Prolog', 'Cobol', 'Apex', 'Crystal', 'Fortran', 'Nim', 'Delphi', 'Ada', 'Objective-C', 'VBA', 'Perl', 'Groovy', 'MATLAB', 'Solidity', 'Visual Basic', 'OCaml', 'Erlang', 'Julia', 'Lisp', 'F#', 'Clojure', 'GDScript', 'Scala', 'R', 'Haskell', 'Ruby', 'Elixir', 'Lua', 'Zig', 'Dart', 'Swift', 'Metal', 'PowerShell', 'PHP', 'Kotlin', 'C', 'Java', 'C++', 'C#', 'Bash/Shell', 'Go', 'Rust', 'TypeScript', 'HTML/CSS', 'SQL', 'JavaScript', 'Python', 'Lean', 'Coq', 'Pony', 'D', 'Racket', 'Haxe', 'x86-64 ASM', 'ARM-64 ASM', 'LLVM IR', 'GLSL', 'CUDA', 'Vulkan'][hash(row|string) % 60] }}\n***\n"
spoken_languages: &spoken_languages "Answer in {{ ['Arabic', 'Chinese', 'French', 'German', 'Hebrew', 'Hindi', 'Japanese', 'Korean', 'Portuguese', 'Russian', 'Spanish', 'Turkish'][hash(row|string) % 12] }}\n***\n"
max_seq_length: 8192
shuffle: true
seed: 42
datasets:
# Category Summary (Total: 590 samples)
# =====================================================
# General chat (24 samples - 4.07%)
# Instruction and Reasoning tuning (14 samples - 2.37%)
# Multilingual (36 samples - 6.10%)
# Tool use (100 samples - 16.95%)
# Code / Programming / Software Engineering / Devops (328 samples - 55.59%)
# Math (12 samples - 2.03%)
# Sciences (16 samples - 2.71%)
# Medical (8 samples - 1.36%)
# Finance (8 samples - 1.36%)
# Business (16 samples - 2.71%)
# Humanities and Philosophy (8 samples - 1.36%)
# Creative Writing, Adventure, Roleplay (13 samples - 2.20%)
# General Knowledge and Pop Culture (2 samples - 0.34%)
# Specialized skills (4 samples - 0.68%)
# Misc (1 sample - 0.17%)
# =====================================================
# Research
# =====================================================
# According to this presentation https://minjiazhang.github.io/courses/fall24-resource/slides/awq.pdf
# AWQ only needs 64 samples to identify salient weights that need to be preserved.
#
# This research predates the boom of MoE (Mixture-of-Experts) models
# and it's safer to assume that 64 samples of a general dataset
# cannot properly identify salient weights of experts.
# General chat (24 samples)
# ---------------------------------------------------------------------------
- dataset: HuggingFaceH4/ultrachat_200k
columns: [messages]
split: train_sft
formatter: chat_completion
num_samples: 8
streaming: true
- dataset: databricks/databricks-dolly-15k
split: train
columns: [instruction, response]
formatter: prompt_answer
num_samples: 8
- dataset: neuralmagic/calibration
subset: LLM
split: train
columns: [messages]
formatter: chat_completion
num_samples: 8
# Instruction and Reasoning tuning (14 samples)
# ---------------------------------------------------------------------------
- dataset: HuggingFaceH4/no_robots
split: train
columns: [messages]
formatter: chat_completion
num_samples: 2
- dataset: nvidia/HelpSteer
split: train
columns: [prompt, response]
formatter: prompt_answer
num_samples: 2
streaming: true
- dataset: garage-bAInd/Open-Platypus
split: train
columns: [instruction, output]
formatter: prompt_answer
num_samples: 2
- dataset: PJMixers/grimulkan_physical-reasoning-ShareGPT
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 4
- dataset: PJMixers/grimulkan_theory-of-mind-ShareGPT
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 4
# Multilingual (36 samples)
# ---------------------------------------------------------------------------
- dataset: HuggingFaceH4/Multilingual-Thinking
split: train
columns: [user]
formatter: raw_text
num_samples: 32
formatter_params:
prefix: *spoken_languages
- dataset: ServiceNow-AI/M2Lingual
subset: full_data
split: train
columns: [conversation]
formatter: chat_completion
num_samples: 4
streaming: true
# Tool use (include commented out ToolAce) (100 samples)
# ---------------------------------------------------------------------------
# Fail with minimax!
# jinja2.exceptions.TemplateError: Message has tool role, but there was no previous assistant message with a tool call!
# - dataset: Team-ACE/ToolACE
# split: train
# columns: [system, conversations]
# formatter: chat_completion_with_sysprompt
# num_samples: 100
- dataset: interstellarninja/hermes_reasoning_tool_use
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 100
streaming: true
# Code / Programming / Software Engineering / Devops (336 samples)
# ---------------------------------------------------------------------------
- dataset: deepmind/code_contests
split: train
columns: [name]
formatter: deepmind_code_contests
num_samples: 50
streaming: true
- dataset: dh02391735/stackoverflow-kubernetes-questions
split: train
columns: [instruction]
formatter: raw_text
num_samples: 8
streaming: true
- dataset: diversoailab/humaneval-rust
split: train
columns: [prompt]
formatter: raw_text
num_samples: 100
formatter_params: # The dataset actually doesn't hardcode the language
prefix: *programming_languages
- dataset: ammarnasr/the-stack-rust-clean
split: train
columns: [content]
formatter: raw_text
num_samples: 8
streaming: true
formatter_params:
prefix: "Explain this code and comment it for a junior dev.\n***\n"
- dataset: CSJianYang/CodeArena
split: test
columns: [messages]
formatter: chat_completion
num_samples: 8
- dataset: nvidia/OpenCodeInstruct
split: train
columns: [input, output]
formatter: prompt_answer
num_samples: 8
streaming: true
- dataset: nvidia/Llama-Nemotron-Post-Training-Dataset
split: code
columns: [input]
formatter: chat_completion
num_samples: 8
streaming: true
- dataset: nvidia/Nemotron-Competitive-Programming-v1
split: competitive_coding_cpp_part00
columns: [messages]
formatter: chat_completion
num_samples: 8
streaming: true
# The conversations columns has another "conversations" field :/
# - dataset: sr5434/CodegebraGPT_data
# subset: 100k-text
# split: train
# columns: [conversations]
# formatter: sharegpt
# num_samples: 8
- dataset: rombodawg/code_bagel_hermes-2.5
split: train
columns: [input, output]
formatter: prompt_answer
num_samples: 100
streaming: true
- dataset: MathArena/project_euler
split: train
columns: [problem]
formatter: raw_text
num_samples: 30
formatter_params:
prefix: *programming_languages
# Math (12 samples)
- dataset: nvidia/Llama-Nemotron-Post-Training-Dataset
split: math
columns: [input]
formatter: chat_completion
num_samples: 4
streaming: true
- dataset: nvidia/Nemotron-Math-Proofs-v1
split: lean
columns: [formal_statement]
formatter: raw_text
num_samples: 4
streaming: true
formatter_params:
prefix: "Can you improve, document and add comment to this Lean proof for a non-mathematician?\n***\n"
- dataset: nvidia/OpenMathInstruct-2
split: train
columns: [problem, generated_solution]
formatter: prompt_answer
num_samples: 4
streaming: true
# Sciences (16 samples)
- dataset: nvidia/Llama-Nemotron-Post-Training-Dataset
split: science
columns: [input]
formatter: chat_completion
num_samples: 4
streaming: true
- dataset: nvidia/OpenScienceReasoning-2
split: train
columns: [input, output]
formatter: prompt_answer
num_samples: 8
streaming: true
- dataset: MegaScience/MegaScience
split: train
columns: [question, answer]
formatter: prompt_answer
num_samples: 4
streaming: true
# Medical (8 samples)
- dataset: OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B
split: train
columns: [messages]
formatter: chat_completion
num_samples: 4
streaming: true
- dataset: ccdv/pubmed-summarization
subset: section
split: train
columns: [article]
formatter: raw_text
num_samples: 4
streaming: true
formatter_params:
prefix: "Summarize this:\n***\n"
# Finance (8 samples)
- dataset: gbharti/finance-alpaca
split: train
columns: [instruction, output]
formatter: prompt_answer
num_samples: 4
- dataset: vladlen32230/summarization-yahoo-stock-finance-article-text
split: train
columns: [text]
formatter: raw_text
num_samples: 4
formatter_params:
prefix: "Summarize this:\n***\n"
# Business (16 samples)
- dataset: fka/awesome-chatgpt-prompts
split: train
columns: [prompt]
formatter: raw_text
num_samples: 8
- dataset: theoldmandthesea/17k_business_book
split: train
columns: [question, answer]
formatter: prompt_answer
num_samples: 8
# Humanities and Philosophy (8 samples)
- dataset: ruggsea/stanford-encyclopedia-of-philosophy_instruct
split: train
columns: [question, answer]
formatter: prompt_answer
num_samples: 2
streaming: true
- dataset: mlfoundations-dev/stackexchange_philosophy
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 2
- dataset: FreedomIntelligence/SocraticChat
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 4
streaming: true
# Creative Writing, Adventure, Roleplay (13 samples)
- dataset: Gryphe/Opus-WritingPrompts
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 2
- dataset: anthracite-org/nopm_claude_writing_fixed
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 2
- dataset: zerofata/Roleplay-Anime-Characters
split: train
columns: [messages]
formatter: chat_completion
num_samples: 1
- dataset: zerofata/Instruct-Anime
split: train
columns: [messages]
formatter: chat_completion
num_samples: 1
- dataset: zerofata/Instruct-Anime-CreativeWriting
split: train
columns: [messages]
formatter: chat_completion
num_samples: 1
- dataset: sam-paech/gutenberg3-generalfiction-scifi-fantasy-romance-adventure-dpo
split: train
columns: [chosen]
formatter: chat_completion
num_samples: 2
- dataset: PocketDoc/Dans-Prosemaxx-Adventure
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 2
- dataset: anthracite-org/stheno-filtered-v1.1
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 2
streaming: true
# General Knowledge and Pop Culture (2 samples)
- dataset: KaraKaraWitch/TvTroper-2025
split: train
columns: [article]
formatter: raw_text
num_samples: 2
streaming: true
formatter_params:
prefix: "Explain this trope like I'm your grandmother\n***\n"
# Behavioral skills (8 samples)
- dataset: AquaV/US-Army-Survival-Sharegpt
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 1
- dataset: AquaV/Interrogation-Sharegpt
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 1
- dataset: AquaV/Multi-Environment-Operations-Sharegpt
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 1
- dataset: AquaV/Resistance-Sharegpt
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 1
# Misc (1 sample)
- dataset: PocketDoc/Dans-Kinomaxx-VanillaBackrooms
split: train
columns: [conversations]
formatter: sharegpt
num_samples: 1
|