add new cms model version, additionally trained on vbf and multi-particle gun

cms/2024_05_16_attn_model21M/README.md

@@ -15,7 +15,7 @@ This model reconstructs particles in a detector, based on the tracks and calorim
 
 <!-- Provide the basic links for the model. -->
 
-- **Repository:** https://github.com/jpata/particleflow/releases/tag/v1.8
+- **Repository:** https://github.com/jpata/particleflow/releases/tag/v1.8.0
 
 ## Uses
 

cms/2024_06_13_attn_model21M/README.md

@@ -0,0 +1 @@
+Fine-tuned from [../2024_05_16_attn_model21M](../2024_05_16_attn_model21M/README.md) for two epochs using additional VBF and multi-particle gun samples.

cms/2024_06_13_attn_model21M/onnx/mlpf_21M_attn2x6x512_bs40_relu_tt_qcd_zh400k_vbf_mpgun_checkpoint27_1xa100_fp32_fused.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1bbcc01ec47c4969b22db439ed24167f0a761bc645279d675a4ec3d63e8dabf
+size 85282119

cms/2024_06_13_attn_model21M/onnx/mlpf_21M_attn2x6x512_bs40_relu_tt_qcd_zh400k_vbf_mpgun_checkpoint27_1xa100_fp32_unfused.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1864c8fe086799213dae7df13209b9ab63be68d0c4e42cdfa6168359d70f047c
+size 85411488

cms/2024_06_13_attn_model21M/pyg-cms_20240430_094836_751206/checkpoints/checkpoint-27-17.613789.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b33fd79e2e511f31c97118f3aaf0dbd5ef3f857b419685cf7d2cdb1ddc781c6
+size 255895346

cms/2024_06_13_attn_model21M/pyg-cms_20240430_094836_751206/history/epoch_26.json

@@ -0,0 +1 @@
+{"train": {"Classification": 11.868144009791035, "Regression": 2.9049221830946013, "MET": 499101.1555233217, "Sliced_Wasserstein_Loss": 2268.2242872036627, "Total": 14.77309080504057, "cls0_Classification": 7.804604987307919, "cls0_Regression": 0.0, "cls1_Classification": 2.1060698886043245, "cls1_Regression": 16.00181315443543, "cls2_Classification": 2.0377357454897784, "cls2_Regression": 57.58269117447079, "cls3_Classification": 3.1613335609219892, "cls3_Regression": 54.23718326458456, "cls4_Classification": 2.821606851457323, "cls4_Regression": 33.65919779248448, "cls5_Classification": 1.0472354175638003, "cls5_Regression": 5.063825869180907, "cls6_Classification": 1.335993275746793, "cls6_Regression": 44.4822877476089}, "valid": {"Classification": 14.635026605714762, "Regression": 2.990642767514995, "MET": 4001.2155948017325, "Sliced_Wasserstein_Loss": 2175.6701099633456, "Total": 17.62565733297234, "cls0_Classification": 7.3277306002165945, "cls0_Regression": 0.0, "cls1_Classification": 1.3040255299014911, "cls1_Regression": 0.26776539981464514, "cls2_Classification": 1.5215393696892703, "cls2_Regression": 4.124176055898034, "cls3_Classification": 3.7881715209513493, "cls3_Regression": 29.0589282739087, "cls4_Classification": 3.3949035217427523, "cls4_Regression": 19.864821101299565, "cls5_Classification": 0.7796326192050567, "cls5_Regression": 0.49206461615659364, "cls6_Classification": 0.11579720618843198, "cls6_Regression": 0.036788081137667655}, "epoch_time": 73547.27697467804}

cms/2024_06_13_attn_model21M/pyg-cms_20240430_094836_751206/history/epoch_27.json

@@ -0,0 +1 @@
+{"train": {"Classification": 11.741978916413581, "Regression": 2.859136138151036, "MET": 479704.38330084766, "Sliced_Wasserstein_Loss": 2279.063143103214, "Total": 14.601080526721363, "cls0_Classification": 7.586577416028285, "cls0_Regression": 0.0, "cls1_Classification": 2.0127830929015005, "cls1_Regression": 14.868471227505554, "cls2_Classification": 1.962295063970355, "cls2_Regression": 54.19230202620008, "cls3_Classification": 3.114653687502833, "cls3_Regression": 52.15139839535833, "cls4_Classification": 2.7836736775304836, "cls4_Regression": 32.77948359095236, "cls5_Classification": 0.9995969095632564, "cls5_Regression": 4.4525328917546805, "cls6_Classification": 1.2358514636972486, "cls6_Regression": 41.971944268165544}, "valid": {"Classification": 14.627535613128957, "Regression": 2.986235707889037, "MET": 4226.481172942353, "Sliced_Wasserstein_Loss": 2190.0763078973678, "Total": 17.613788893285573, "cls0_Classification": 7.327428617544152, "cls0_Regression": 0.0, "cls1_Classification": 1.3200939042454598, "cls1_Regression": 0.261243892009955, "cls2_Classification": 1.4920587994106131, "cls2_Regression": 4.127097933501333, "cls3_Classification": 3.881660214824225, "cls3_Regression": 29.036042673275574, "cls4_Classification": 3.3314256575933023, "cls4_Regression": 19.832908561729422, "cls5_Classification": 0.7517686161904366, "cls5_Regression": 0.4908277644431023, "cls6_Classification": 0.11740967529052816, "cls6_Regression": 0.03632755908756326}, "epoch_time": 73389.73045301437}

cms/2024_06_13_attn_model21M/pyg-cms_20240430_094836_751206/model_kwargs.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89e133f916282d927422d3cf5834d89b4e9abffb5086f24792b90b0a3b49c257
+size 514

cms/2024_06_13_attn_model21M/pyg-cms_20240430_094836_751206/overridden_config.yaml

@@ -0,0 +1,130 @@
+backend: pytorch
+checkpoint_freq: 1
+comet: true
+comet_name: particleflow-pt
+comet_offline: false
+comet_step_freq: 10
+config: parameters/pytorch/pyg-cms.yaml
+conv_type: attention
+data_dir: /scratch/persistent/joosep/tensorflow_datasets
+dataset: cms
+dtype: bfloat16
+gpu_batch_multiplier: 40
+gpus: 1
+load: experiments/pyg-cms_20240430_094836_751206/checkpoints/checkpoint-25-17.631161.pth
+lr: 5.0e-05
+lr_schedule: cosinedecay
+lr_schedule_config:
+  onecycle:
+    pct_start: 0.3
+model:
+  attention:
+    activation: relu
+    attention_type: flash
+    conv_type: attention
+    dropout_conv_id_ff: 0.0
+    dropout_conv_id_mha: 0.0
+    dropout_conv_reg_ff: 0.0
+    dropout_conv_reg_mha: 0.0
+    dropout_ff: 0.0
+    head_dim: 16
+    num_convs: 6
+    num_heads: 32
+  cos_phi_mode: linear
+  energy_mode: linear
+  eta_mode: linear
+  gnn_lsh:
+    activation: elu
+    bin_size: 640
+    conv_type: gnn_lsh
+    distance_dim: 128
+    dropout_ff: 0.0
+    embedding_dim: 512
+    ffn_dist_hidden_dim: 128
+    ffn_dist_num_layers: 2
+    layernorm: true
+    max_num_bins: 200
+    num_convs: 3
+    num_node_messages: 2
+    width: 512
+  input_encoding: joint
+  learned_representation_mode: last
+  mamba:
+    activation: elu
+    conv_type: mamba
+    d_conv: 4
+    d_state: 32
+    dropout_ff: 0.0
+    embedding_dim: 1024
+    expand: 2
+    num_convs: 4
+    width: 1024
+  pt_mode: linear
+  sin_phi_mode: linear
+  trainable: all
+ntest: null
+ntrain: null
+num_epochs: 100
+num_workers: 4
+nvalid: null
+patience: 20
+prefetch_factor: 50
+ray_train: false
+raytune:
+  asha:
+    brackets: 1
+    grace_period: 10
+    max_t: 200
+    reduction_factor: 4
+  default_metric: val_loss
+  default_mode: min
+  hyperband:
+    max_t: 200
+    reduction_factor: 4
+  hyperopt:
+    n_random_steps: 10
+  local_dir: null
+  nevergrad:
+    n_random_steps: 10
+  sched: asha
+  search_alg: hyperopt
+sort_data: true
+test_dataset:
+  cms_pf_qcd:
+    version: 1.7.1
+  cms_pf_ttbar:
+    version: 1.7.1
+  cms_pf_ztt:
+    version: 1.7.1
+test_datasets: []
+train: true
+train_dataset:
+  cms:
+    gun:
+      batch_size: 5
+      samples:
+        cms_pf_multi_particle_gun:
+          version: 1.7.1
+    physical:
+      batch_size: 1
+      samples:
+        cms_pf_qcd:
+          version: 1.7.1
+        cms_pf_ttbar:
+          version: 1.7.1
+        cms_pf_vbf:
+          version: 1.7.1
+        cms_pf_ztt:
+          version: 1.7.1
+val_freq: null
+valid_dataset:
+  cms:
+    physical:
+      batch_size: 1
+      samples:
+        cms_pf_qcd:
+          version: 1.7.1
+        cms_pf_ttbar:
+          version: 1.7.1
+        cms_pf_ztt:
+          version: 1.7.1