Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77363.10328603323,
-      "required_pages_per_second_upper_estimate": 0.005352158992458659
+      "required_token_positions_per_second": 76410.49206095206,
+      "required_pages_per_second_upper_estimate": 0.005286255137544973
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9670.387910754154,
-      "required_pages_per_second_upper_estimate": 0.00404878947756975
+      "required_token_positions_per_second": 9551.311507619008,
+      "required_pages_per_second_upper_estimate": 0.003998934674175122
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19340.77582150831,
-      "required_pages_per_second_upper_estimate": 0.004028173553062353
+      "required_token_positions_per_second": 19102.623015238016,
+      "required_pages_per_second_upper_estimate": 0.003978572604028099
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29011.163732262463,
-      "required_pages_per_second_upper_estimate": 0.001739174968267023
+      "required_token_positions_per_second": 28653.934522857024,
+      "required_pages_per_second_upper_estimate": 0.0017177596226205865
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29011.163732262463,
-      "required_pages_per_second_upper_estimate": 0.002255490373560032
+      "required_token_positions_per_second": 28653.934522857024,
+      "required_pages_per_second_upper_estimate": 0.0022277173738139925
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29011.163732262463,
-      "required_pages_per_second_upper_estimate": 0.37212405859676584
+      "required_token_positions_per_second": 28653.934522857024,
+      "required_pages_per_second_upper_estimate": 0.36754190586135405
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.29443341789847466,
-    "producer_p95_seconds": 0.22574543500013533,
-    "gpu_update_seconds": 0.9530124422155969,
-    "producer_batches_per_second": 3.3963536039404874,
-    "gpu_batches_per_second": 1.0493042437884281,
-    "producer_headroom": 3.236767242719068,
+    "producer_batch_seconds": 0.08254387310951472,
+    "producer_p95_seconds": 0.22626898999988043,
+    "gpu_update_seconds": 0.9648936685447299,
+    "producer_batches_per_second": 12.114769544109645,
+    "gpu_batches_per_second": 1.0363836271287987,
+    "producer_headroom": 11.68946442898992,
     "sustainable": true,
-    "buffer_seconds": 60.039783859582606,
+    "buffer_seconds": 60.78830111831798,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 511917.20838742284,
+    "generation_bytes_per_second": 529541.4258084036,
     "upload_bytes_per_second": 5454454.944979651,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.84551609723104,
+    "estimated_full_shards_per_hour": 113.62726288498956,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 774900, "loss": 0.5177415013313293, "lr": 3e-05, "grad_norm": 0.25173185375294777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113423131229, "tokens_per_second": 151947.70477963556, "data_wait_s": 1.499599966336973e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775000, "loss": 0.49935993552207947, "lr": 3e-05, "grad_norm": 0.2471969544188181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113437606124, "tokens_per_second": 150626.16408695595, "data_wait_s": 1.0480000128154643e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775100, "loss": 0.5219948291778564, "lr": 3e-05, "grad_norm": 0.2515800013685113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113452290848, "tokens_per_second": 141605.58837923338, "data_wait_s": 1.4733999705640599e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 775200, "loss": 0.5444227457046509, "lr": 3e-05, "grad_norm": 0.25118968262756286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113466870528, "tokens_per_second": 151271.90771167693, "data_wait_s": 1.2540000170702115e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2ee210e96ee5ea2342fa91785f20ef7745ffbf805738023f5ff02857e187c254
-size 322332
+oid sha256:f48ef3360bdbd8243b4582470991e6ae57f02dd39e3a198de34919785d926eec
+size 324450