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": 76996.56683626222,
-      "required_pages_per_second_upper_estimate": 0.005326801150381763
+      "required_token_positions_per_second": 76670.2665246559,
+      "required_pages_per_second_upper_estimate": 0.005304226937703805
     },
     {
       "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": 9624.570854532778,
-      "required_pages_per_second_upper_estimate": 0.004029606832898791
+      "required_token_positions_per_second": 9583.783315581988,
+      "required_pages_per_second_upper_estimate": 0.004012529942080763
     },
     {
       "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": 19249.141709065556,
-      "required_pages_per_second_upper_estimate": 0.004009088583994579
+      "required_token_positions_per_second": 19167.566631163976,
+      "required_pages_per_second_upper_estimate": 0.003992098646547151
     },
     {
       "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": 28873.712563598332,
-      "required_pages_per_second_upper_estimate": 0.0017309349805813906
+      "required_token_positions_per_second": 28751.349946745962,
+      "required_pages_per_second_upper_estimate": 0.0017235995285379887
     },
     {
       "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": 28873.712563598332,
-      "required_pages_per_second_upper_estimate": 0.002244804149780192
+      "required_token_positions_per_second": 28751.349946745962,
+      "required_pages_per_second_upper_estimate": 0.0022352909945361917
     },
     {
       "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": 28873.712563598332,
-      "required_pages_per_second_upper_estimate": 0.37036098258870886
+      "required_token_positions_per_second": 28751.349946745962,
+      "required_pages_per_second_upper_estimate": 0.36879144632246846
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0809062593749843,
-    "producer_p95_seconds": 0.18695144800039998,
-    "gpu_update_seconds": 0.9575491873135977,
-    "producer_batches_per_second": 12.359983117810458,
-    "gpu_batches_per_second": 1.0443327750144074,
-    "producer_headroom": 11.83529178966919,
+    "producer_batch_seconds": 0.07530499859379347,
+    "producer_p95_seconds": 0.18405966800128226,
+    "gpu_update_seconds": 0.9616244124609934,
+    "producer_batches_per_second": 13.279330969702967,
+    "gpu_batches_per_second": 1.0399070437914484,
+    "producer_headroom": 12.76972884161569,
     "sustainable": true,
-    "buffer_seconds": 60.32559880075666,
+    "buffer_seconds": 60.582337985042585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337550.95430779195,
-    "upload_bytes_per_second": 5952811.925259532,
+    "generation_bytes_per_second": 337666.61801866704,
+    "upload_bytes_per_second": 5950355.542605752,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.43057701039618,
+    "estimated_full_shards_per_hour": 72.45539575023659,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 11547889,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.05019463699864,
+      "bytes": 6120696,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 490900, "loss": 0.6020042300224304, "lr": 0.0003, "grad_norm": 0.19061114434149673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71851442957, "tokens_per_second": 116923.8574456793, "data_wait_s": 1.2448999768821523e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491000, "loss": 0.5649935603141785, "lr": 0.0003, "grad_norm": 0.16114400338628823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71865918152, "tokens_per_second": 150880.5116661854, "data_wait_s": 9.368999599246308e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 491100, "loss": 0.625411868095398, "lr": 0.0003, "grad_norm": 0.19665559218478687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71880471324, "tokens_per_second": 151413.50245062984, "data_wait_s": 1.2169000910944305e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 491200, "loss": 0.5591356754302979, "lr": 0.0003, "grad_norm": 0.14437696167595815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71894998167, "tokens_per_second": 151455.14696415327, "data_wait_s": 1.0092000593431294e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a5b0d82fa365eef53a61596d855d847f89466ee93f97e32aeef3a2236f4a489
-size 130146
+oid sha256:5377f5956ded944d25e49e848b6bc1d5d5701c756dec6e5777010161987e826d
+size 132264