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": 77200.05601747986,
-      "required_pages_per_second_upper_estimate": 0.005340879004098351
+      "required_token_positions_per_second": 77167.66187820998,
+      "required_pages_per_second_upper_estimate": 0.005338637902378902
     },
     {
       "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": 9650.007002184982,
-      "required_pages_per_second_upper_estimate": 0.004040256416753605
+      "required_token_positions_per_second": 9645.957734776248,
+      "required_pages_per_second_upper_estimate": 0.004038561073047887
     },
     {
       "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": 19300.014004369965,
-      "required_pages_per_second_upper_estimate": 0.004019683941513842
+      "required_token_positions_per_second": 19291.915469552496,
+      "required_pages_per_second_upper_estimate": 0.004017997230284088
     },
     {
       "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": 28950.02100655495,
-      "required_pages_per_second_upper_estimate": 0.001735509555220394
+      "required_token_positions_per_second": 28937.873204328742,
+      "required_pages_per_second_upper_estimate": 0.0017347813130255529
     },
     {
       "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": 28950.02100655495,
-      "required_pages_per_second_upper_estimate": 0.002250736795574701
+      "required_token_positions_per_second": 28937.873204328742,
+      "required_pages_per_second_upper_estimate": 0.0022497923573841483
     },
     {
       "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": 28950.02100655495,
-      "required_pages_per_second_upper_estimate": 0.3713397853613339
+      "required_token_positions_per_second": 28937.873204328742,
+      "required_pages_per_second_upper_estimate": 0.3711839663976699
     }
   ],
   "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": 0.0890228193517828,
-    "producer_p95_seconds": 0.3273739869982819,
-    "gpu_update_seconds": 0.9550252137551077,
-    "producer_batches_per_second": 11.233074927097034,
-    "gpu_batches_per_second": 1.0470927736745859,
-    "producer_headroom": 10.727869783377987,
+    "producer_batch_seconds": 0.14941703393759553,
+    "producer_p95_seconds": 0.3312021140009165,
+    "gpu_update_seconds": 0.9554261228798323,
+    "producer_batches_per_second": 6.69267735844397,
+    "gpu_batches_per_second": 1.0466534000408254,
+    "producer_headroom": 6.394358780263761,
     "sustainable": true,
-    "buffer_seconds": 60.16658846657179,
+    "buffer_seconds": 60.19184574142943,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343830.01957017544,
+    "generation_bytes_per_second": 342983.2959457252,
     "upload_bytes_per_second": 4859119.183575832,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.77791824654528,
+    "estimated_full_shards_per_hour": 73.59623106745546,
     "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

@@ -359,3 +359,4 @@
 {"step": 795900, "loss": 0.5025835633277893, "lr": 3e-05, "grad_norm": 0.2636277595490215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116497429120, "tokens_per_second": 152916.09176264008, "data_wait_s": 8.536000677850097e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 796000, "loss": 0.5353186130523682, "lr": 3e-05, "grad_norm": 0.2668817380587538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116511894043, "tokens_per_second": 150692.8289614042, "data_wait_s": 8.601993613410741e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 796100, "loss": 0.5252862572669983, "lr": 3e-05, "grad_norm": 0.27843552186022336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116526403927, "tokens_per_second": 150740.62097914968, "data_wait_s": 1.095300103770569e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 796200, "loss": 0.5428974628448486, "lr": 3e-05, "grad_norm": 0.2584630178822616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116541001292, "tokens_per_second": 151880.1621858557, "data_wait_s": 1.971399615285918e-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 sha256:722a6d0494a56800f7866eb76d3cc0eb30ab01af856e08091ae64e37ead8dba5
-size 770244
+oid sha256:1d9d4ec1a935da521d88433d8e57bee957057f0e6121776cf7b0939b8a334971
+size 772362