Skip to content

[DRAFT] Speedrun Submission nanogpt_features_v0 #2184

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

Draft
ClassicLarry wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

[DRAFT] Speedrun Submission nanogpt_features_v0 #2184

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8587a43
nanogpt_features_v0
ClassicLarry Dec 7, 2025
4b3c38d
Merge branch 'main' into nanogpt_features_v0
Helw150 Dec 12, 2025
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
139 changes: 139 additions & 0 deletions experiments/speedrun/nanogpt_features_v0/150m/speedrun_results.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,139 @@
{
"runs": [
{
"run_info": {
"author": {
"affiliation": "Independent",
"name": "Larry Dial",
"url": "https://github.com/ClassicLarry"
},
"description": "Includes subset of features from Modded-NanoGPT: Partial RoPE, QK Norm, 2.5 TPP, Relu^2 MLP, X0 Skip, exponential decay of resid, backout lambda, reduced head counts, rms_norm, 0 init out projections, boosted attn scale (150m)",
"device_flops": 989500000000000.0,
"eval/paloma/c4_en/bpb": 1.2709236145019531,
"model_config": {
"attn_backend": "JAX_FLASH",
"cross_entropy_block_size": 2048,
"flash_attention_block_size": null,
"gradient_checkpointing": true,
"head_dim": null,
"hidden_dim": 512,
"initializer_range": 0.02,
"input_embedding_norm": true,
"intermediate_dim": 2048,
"layer_norm_epsilon": 1e-05,
"num_heads": 4,
"num_kv_heads": 4,
"num_layers": 6,
"qk_norm": {
"eps": 1e-05,
"use_bias": false,
"use_weight": false
},
"reference_checkpoint": "NousResearch/Llama-2-7b-hf",
"seq_len": 2048,
"tie_word_embeddings": false,
"tokenizer": null,
"upcast_attn": false,
"use_attention_sink": false,
"use_bias": false,
"use_layer_norm_weight": false
},
"model_flops": 2.189996844908544e+17,
"model_flops_per_token": 194396160.0,
"model_size": 150208512,
"num_chips": 1,
"num_devices": 1,
"resources": {
"cpu": 1,
"device": {
"count": 1,
"type": "H100"
},
"disk": "1g",
"preemptible": true,
"ram": "128m",
"regions": null,
"replicas": 1
},
"run_completion_timestamp": "2025-12-07 19:46:51 UTC",
"tokenized_dataset": "/root/marin/local_store/tokenized/subcache/fineweb-edu-10B-ac65f6",
"total_tokens": 375521280,
"train_config": {
"allow_partial_checkpoint": false,
"beta1": null,
"beta2": null,
"cycle_length": null,
"data_seed": null,
"decay": null,
"ema_beta": null,
"epsilon": null,
"initialize_from_checkpoint_path": null,
"initialize_from_hf": null,
"int8": false,
"learning_rate": 0.02,
"lr_schedule": null,
"max_eval_batches": null,
"max_grad_norm": null,
"min_lr_ratio": null,
"num_train_steps": 955,
"optimizer_config": {
"adam_lr": 0.0064,
"adam_weight_decay": null,
"backend_steps": 5,
"beta1": 0.8,
"beta2": 0.95,
"cooldown": null,
"cycle_length": null,
"cycles": null,
"decay": 0.5,
"default_weight_decay_mask": null,
"epsilon": 1e-15,
"haps": null,
"learning_rate": 0.02,
"lr": 0.02,
"lr_schedule": "linear",
"max_grad_norm": 1,
"min_lr_ratio": 0.1,
"momentum": 0.95,
"muon_epsilon": 1e-05,
"nesterov": true,
"rewarmup": 0.0,
"use_kimi_scaling": false,
"warmup": 0,
"weight_decay": 0,
"weight_decay_modules": null
},
"per_device_eval_parallelism": null,
"profiler": false,
"profiler_num_steps": 100,
"profiler_start_step": 5,
"reset_data_loader_on_init": true,
"rewarmup": null,
"skip_bad_steps": false,
"steps_per_eval": null,
"steps_per_export": 10000,
"steps_per_hf_export": -1,
"steps_per_task_eval": null,
"train_batch_size": 192,
"warmup": null,
"watch": {
"include_histograms": false,
"include_norms": true,
"include_per_parameter_norms": true,
"interval": 10,
"split_scan_layers": true,
"watch_targets": [
"grads",
"params"
]
},
"weight_decay": null,
"z_loss_weight": null
},
"training_hardware_flops": 1.1208567897015704e+18,
"training_time": 1132.750671754998,
"wandb_run_link": "https://wandb.ai/larrydial/marin/runs/nanogpt_features_v0_150m-5c83f9"
}
}
]
}
139 changes: 139 additions & 0 deletions experiments/speedrun/nanogpt_features_v0/270m/speedrun_results.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,139 @@
{
"runs": [
{
"run_info": {
"author": {
"affiliation": "Independent",
"name": "Larry Dial",
"url": "https://github.com/ClassicLarry"
},
"description": "Includes subset of features from Modded-NanoGPT: Partial RoPE, QK Norm, 2.5 TPP, Relu^2 MLP, X0 Skip, exponential decay of resid, backout lambda, reduced head counts, rms_norm, 0 init out projections, boosted attn scale (270m)",
"device_flops": 989500000000000.0,
"eval/paloma/c4_en/bpb": 1.1750104427337646,
"model_config": {
"attn_backend": "JAX_FLASH",
"cross_entropy_block_size": 2048,
"flash_attention_block_size": null,
"gradient_checkpointing": true,
"head_dim": null,
"hidden_dim": 768,
"initializer_range": 0.02,
"input_embedding_norm": true,
"intermediate_dim": 3072,
"layer_norm_epsilon": 1e-05,
"num_heads": 6,
"num_kv_heads": 6,
"num_layers": 11,
"qk_norm": {
"eps": 1e-05,
"use_bias": false,
"use_weight": false
},
"reference_checkpoint": "NousResearch/Llama-2-7b-hf",
"seq_len": 2048,
"tie_word_embeddings": false,
"tokenizer": null,
"upcast_attn": false,
"use_attention_sink": false,
"use_bias": false,
"use_layer_norm_weight": false
},
"model_flops": 8.703490096701112e+17,
"model_flops_per_token": 422326272.0,
"model_size": 274857984,
"num_chips": 1,
"num_devices": 1,
"resources": {
"cpu": 1,
"device": {
"count": 1,
"type": "H100"
},
"disk": "1g",
"preemptible": true,
"ram": "128m",
"regions": null,
"replicas": 1
},
"run_completion_timestamp": "2025-12-07 20:38:36 UTC",
"tokenized_dataset": "/root/marin/local_store/tokenized/subcache/fineweb-edu-10B-ac65f6",
"total_tokens": 686948352,
"train_config": {
"allow_partial_checkpoint": false,
"beta1": null,
"beta2": null,
"cycle_length": null,
"data_seed": null,
"decay": null,
"ema_beta": null,
"epsilon": null,
"initialize_from_checkpoint_path": null,
"initialize_from_hf": null,
"int8": false,
"learning_rate": 0.02,
"lr_schedule": null,
"max_eval_batches": null,
"max_grad_norm": null,
"min_lr_ratio": null,
"num_train_steps": 1747,
"optimizer_config": {
"adam_lr": 0.0064,
"adam_weight_decay": null,
"backend_steps": 5,
"beta1": 0.8,
"beta2": 0.95,
"cooldown": null,
"cycle_length": null,
"cycles": null,
"decay": 0.5,
"default_weight_decay_mask": null,
"epsilon": 1e-15,
"haps": null,
"learning_rate": 0.02,
"lr": 0.02,
"lr_schedule": "linear",
"max_grad_norm": 1,
"min_lr_ratio": 0.1,
"momentum": 0.95,
"muon_epsilon": 1e-05,
"nesterov": true,
"rewarmup": 0.0,
"use_kimi_scaling": false,
"warmup": 0,
"weight_decay": 0,
"weight_decay_modules": null
},
"per_device_eval_parallelism": null,
"profiler": false,
"profiler_num_steps": 100,
"profiler_start_step": 5,
"reset_data_loader_on_init": true,
"rewarmup": null,
"skip_bad_steps": false,
"steps_per_eval": null,
"steps_per_export": 10000,
"steps_per_hf_export": -1,
"steps_per_task_eval": null,
"train_batch_size": 192,
"warmup": null,
"watch": {
"include_histograms": false,
"include_norms": true,
"include_per_parameter_norms": true,
"interval": 10,
"split_scan_layers": true,
"watch_targets": [
"grads",
"params"
]
},
"weight_decay": null,
"z_loss_weight": null
},
"training_hardware_flops": 3.918996717548913e+18,
"training_time": 3960.5828373409936,
"wandb_run_link": "https://wandb.ai/larrydial/marin/runs/nanogpt_features_v0_270m-21987d"
}
}
]
}
Loading
Loading