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": 77183.83364388195,
-      "required_pages_per_second_upper_estimate": 0.005339756702651755
+      "required_token_positions_per_second": 77191.67697692673,
+      "required_pages_per_second_upper_estimate": 0.005340299322630832
     },
     {
       "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": 9647.979205485244,
-      "required_pages_per_second_upper_estimate": 0.004039407420620635
+      "required_token_positions_per_second": 9648.95962211584,
+      "required_pages_per_second_upper_estimate": 0.004039817900590438
     },
     {
       "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": 19295.95841097049,
-      "required_pages_per_second_upper_estimate": 0.004018839268361921
+      "required_token_positions_per_second": 19297.91924423168,
+      "required_pages_per_second_upper_estimate": 0.004019247658219577
     },
     {
       "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": 28943.937616455736,
-      "required_pages_per_second_upper_estimate": 0.001735144865272745
+      "required_token_positions_per_second": 28946.87886634752,
+      "required_pages_per_second_upper_estimate": 0.0017353211887127288
     },
     {
       "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": 28943.937616455736,
-      "required_pages_per_second_upper_estimate": 0.002250263838752492
+      "required_token_positions_per_second": 28946.87886634752,
+      "required_pages_per_second_upper_estimate": 0.0022504925079943876
     },
     {
       "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": 28943.937616455736,
-      "required_pages_per_second_upper_estimate": 0.37126175416497653
+      "required_token_positions_per_second": 28946.87886634752,
+      "required_pages_per_second_upper_estimate": 0.37129948136052027
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09582371539841006,
-    "producer_p95_seconds": 0.2826424129998486,
-    "gpu_update_seconds": 0.9552259394133387,
-    "producer_batches_per_second": 10.435829959653102,
-    "gpu_batches_per_second": 1.0468727436507428,
-    "producer_headroom": 9.9685754767675,
+    "producer_batch_seconds": 0.0998345161719385,
+    "producer_p95_seconds": 0.2888316950011358,
+    "gpu_update_seconds": 0.9551288803071082,
+    "producer_batches_per_second": 10.016575813096194,
+    "gpu_batches_per_second": 1.0469791256636112,
+    "producer_headroom": 9.567120840873828,
     "sustainable": true,
-    "buffer_seconds": 60.179234183040336,
+    "buffer_seconds": 60.17311945934782,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361022.57608475303,
-    "upload_bytes_per_second": 6192202.297810245,
+    "generation_bytes_per_second": 375709.0792241682,
+    "upload_bytes_per_second": 6181980.00635588,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.4670406523413,
+    "estimated_full_shards_per_hour": 80.6184223417643,
     "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": [
-    {
-      "seconds": 1.0359600679985306,
-      "bytes": 6649243,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9754138270000112,
       "bytes": 5934822,
@@ -562,6 +556,12 @@
       "bytes": 17665783,
       "pages": 34,
       "shards": 5
+    },
+    {
+      "seconds": 1.0921460530007607,
+      "bytes": 6276750,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -284,3 +284,4 @@
 {"step": 513400, "loss": 0.5800940990447998, "lr": 0.0003, "grad_norm": 0.14829326745810398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75144253114, "tokens_per_second": 151573.63646637628, "data_wait_s": 8.833998435875401e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 513500, "loss": 0.5706549286842346, "lr": 0.0003, "grad_norm": 0.19791641517325062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75158990870, "tokens_per_second": 126498.12439472727, "data_wait_s": 1.5369998436653987e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 513600, "loss": 0.6071254014968872, "lr": 0.0003, "grad_norm": 0.1688218516042935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75173572909, "tokens_per_second": 151571.5938336054, "data_wait_s": 1.671099744271487e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 513700, "loss": 0.5523330569267273, "lr": 0.0003, "grad_norm": 0.17923486800237254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75188151363, "tokens_per_second": 151950.55079426762, "data_wait_s": 2.088300243485719e-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:cfa6c4d184af0de04b02e9cce2437ec0ac8aef9bac6eb608b030bfdb92cefdeb
-size 609828
+oid sha256:cfe4a2903eed9476d3cc5dd74ad376910edb7f9f07453770e2802d7a59c8b96a
+size 611946