kobzaond/RLVRAMBench
RLVRAMBench Which language-model training configurations can I use with the memory I have, and how much testing does that decision require? RLVRAMBench is a measurement dataset with open evaluation tasks for a specific language-model training system. It measures memory feasibility when response generation and reinforcement-learning updates share the same graphics processors. It provides measured outcomes, fixed prediction tasks, a budgeted decision replay, reference methods, and… See the full description on the dataset page: https://huggingface.co/datasets/kobzaond/RLVRAMBench.
0223
1{2 "Qwen/Qwen2.5-3B-Instruct": {3 "all_linear_layer_adapter_elements": 119734272,4 "config_sha256": "eed00b17e22553979d090fa492e587e92885e328914c8e0b0b78f0a0d3576b3b",5 "floating_checkpoint_bytes": 6171877376,6 "floating_checkpoint_elements": 3085938688,7 "head_dim": 128,8 "hidden_size": 2048,9 "intermediate_size": 11008,10 "largest_transformer_layer_elements": 77076992,11 "largest_weight_tensor_elements": 311164928,12 "lora_rank": 64,13 "model_id": "Qwen/Qwen2.5-3B-Instruct",14 "non_transformer_layer_elements": 311166976,15 "num_attention_heads": 16,16 "num_hidden_layers": 36,17 "num_key_value_heads": 2,18 "revision": "aa8e72537993ba99e69dfaafa59ed015b17504d1",19 "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.",20 "shard_header_sha256": {21 "model-00001-of-00002.safetensors": "8d1c57f12799cf1a917776014a04eff3bfaf1cb1dafa2ca3a12f92ade34e65bd",22 "model-00002-of-00002.safetensors": "5bb7481ec0a898ebaa815dbdd881d83752dfc05e445e19a7571800dbd025cf8e"23 },24 "tensor_count": 434,25 "tie_word_embeddings": true,26 "vocab_size": 15193627 },28 "Qwen/Qwen2.5-7B-Instruct": {29 "all_linear_layer_adapter_elements": 161480704,30 "config_sha256": "7463bb0ea78315365e6c6b74de4e73bbcc8359dfb0c5a737584e077d42c0b03c",31 "floating_checkpoint_bytes": 15231233024,32 "floating_checkpoint_elements": 7615616512,33 "head_dim": 128,34 "hidden_size": 3584,35 "intermediate_size": 18944,36 "largest_transformer_layer_elements": 233057792,37 "largest_weight_tensor_elements": 544997376,38 "lora_rank": 64,39 "model_id": "Qwen/Qwen2.5-7B-Instruct",40 "non_transformer_layer_elements": 1089998336,41 "num_attention_heads": 28,42 "num_hidden_layers": 28,43 "num_key_value_heads": 4,44 "revision": "a09a35458c702b33eeacc393d103063234e8bc28",45 "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.",46 "shard_header_sha256": {47 "model-00001-of-00004.safetensors": "24ce65896dba011b065dd8fcbf877559a6701fe80516f7d4d039d082147cb2e8",48 "model-00002-of-00004.safetensors": "69f0c2c87ae06b271181d6964cf235aa9ce7931f64cde7a1a5511ba16070c739",49 "model-00003-of-00004.safetensors": "a859e7527f6cf5f144667f65c356c01374f95ed56f40cbe6dca4cb7752164066",50 "model-00004-of-00004.safetensors": "181209f962ac9f71cb2bf72adbd5727cf4fbbd646cf80558576ccb5c13e75b71"51 },52 "tensor_count": 339,53 "tie_word_embeddings": false,54 "vocab_size": 15206455 },56 "ibm-granite/granite-3.3-2b-instruct": {57 "all_linear_layer_adapter_elements": 112721920,58 "config_sha256": "9202d328d8368958aab7dad89a8e6aa35c250b3a2eedd2d8850ee0bcceca65f6",59 "floating_checkpoint_bytes": 5067079680,60 "floating_checkpoint_elements": 2533539840,61 "head_dim": 64,62 "hidden_size": 2048,63 "intermediate_size": 8192,64 "largest_transformer_layer_elements": 60821504,65 "largest_weight_tensor_elements": 100677632,66 "lora_rank": 64,67 "model_id": "ibm-granite/granite-3.3-2b-instruct",68 "non_transformer_layer_elements": 100679680,69 "num_attention_heads": 32,70 "num_hidden_layers": 40,71 "num_key_value_heads": 8,72 "revision": "707f574c62054322f6b5b04b6d075f0a8f05e0f0",73 "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.",74 "shard_header_sha256": {75 "model-00001-of-00002.safetensors": "744c520c1a1d794563fed804e54d99dd424713d4346c3edc06dfdd17c6afe1c0",76 "model-00002-of-00002.safetensors": "039f729e7d783e8514dcac3af74ef1e9ad7c8c235e7d92671000e778ab937d79"77 },78 "tensor_count": 362,79 "tie_word_embeddings": true,80 "vocab_size": 4915981 },82 "microsoft/Phi-4-mini-instruct": {83 "all_linear_layer_adapter_elements": 92274688,84 "config_sha256": "ac65d86061d3d0d704ee2511fd0eb8713ef19eb6eedba17c3080a4165d5b933b",85 "floating_checkpoint_bytes": 7672043520,86 "floating_checkpoint_elements": 3836021760,87 "head_dim": 128,88 "hidden_size": 3072,89 "intermediate_size": 8192,90 "largest_transformer_layer_elements": 100669440,91 "largest_weight_tensor_elements": 614596608,92 "lora_rank": 64,93 "model_id": "microsoft/Phi-4-mini-instruct",94 "non_transformer_layer_elements": 614599680,95 "num_attention_heads": 24,96 "num_hidden_layers": 32,97 "num_key_value_heads": 8,98 "revision": "cfbefacb99257ffa30c83adab238a50856ac3083",99 "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.",100 "shard_header_sha256": {101 "model-00001-of-00002.safetensors": "29dd019fbd86197c6d12794e36555ab5bbd74a3b0acfe7c0736176bbea03a2ca",102 "model-00002-of-00002.safetensors": "54aed6db58d46324a40d42491b03f1e6d0c06da5022d61457d992c4f570d4c41"103 },104 "tensor_count": 194,105 "tie_word_embeddings": true,106 "vocab_size": 200064107 }108}109