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": 77194.31955515126,
-      "required_pages_per_second_upper_estimate": 0.0053404821423240325
+      "required_token_positions_per_second": 76954.89562892317,
+      "required_pages_per_second_upper_estimate": 0.005323918239567526
     },
     {
       "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": 9649.289944393908,
-      "required_pages_per_second_upper_estimate": 0.004039956199630351
+      "required_token_positions_per_second": 9619.361953615396,
+      "required_pages_per_second_upper_estimate": 0.004027425974858904
     },
     {
       "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": 19298.579888787815,
-      "required_pages_per_second_upper_estimate": 0.004019385253058252
+      "required_token_positions_per_second": 19238.723907230793,
+      "required_pages_per_second_upper_estimate": 0.004006918830608308
     },
     {
       "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": 28947.869833181725,
-      "required_pages_per_second_upper_estimate": 0.0017353805956613207
+      "required_token_positions_per_second": 28858.08586084619,
+      "required_pages_per_second_upper_estimate": 0.0017299981836119932
     },
     {
       "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": 28947.869833181725,
-      "required_pages_per_second_upper_estimate": 0.0022505695513069445
+      "required_token_positions_per_second": 28858.08586084619,
+      "required_pages_per_second_upper_estimate": 0.0022435892423758146
     },
     {
       "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": 28947.869833181725,
-      "required_pages_per_second_upper_estimate": 0.37131219241905217
+      "required_token_positions_per_second": 28858.08586084619,
+      "required_pages_per_second_upper_estimate": 0.37016054002445054
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1421929561799402,
-    "producer_p95_seconds": 0.3249501840036828,
-    "gpu_update_seconds": 0.9550961835647147,
-    "producer_batches_per_second": 7.032697166338781,
-    "gpu_batches_per_second": 1.0470149679246863,
-    "producer_headroom": 6.716902223736554,
+    "producer_batch_seconds": 0.09519387139062019,
+    "producer_p95_seconds": 0.3425569929968333,
+    "gpu_update_seconds": 0.9580677018329896,
+    "producer_batches_per_second": 10.504877944259485,
+    "gpu_batches_per_second": 1.0437675730919485,
+    "producer_headroom": 10.064384270092745,
     "sustainable": true,
-    "buffer_seconds": 60.17105956457703,
+    "buffer_seconds": 60.35826521547835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333714.8973884138,
-    "upload_bytes_per_second": 5378697.241038193,
+    "generation_bytes_per_second": 320318.0529065974,
+    "upload_bytes_per_second": 5435071.326321987,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.60744849433242,
+    "estimated_full_shards_per_hour": 68.73279753111306,
     "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": 1.1497695119978744,
-      "bytes": 863102,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.423836862002645,
       "bytes": 10998784,
@@ -562,6 +556,12 @@
       "bytes": 7101020,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.235225882002851,
+      "bytes": 5723234,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -345,3 +345,4 @@
 {"step": 719500, "loss": 0.5306189656257629, "lr": 8e-05, "grad_norm": 0.20661925931731365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105314809184, "tokens_per_second": 152209.82406435363, "data_wait_s": 1.556299685034901e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 719600, "loss": 0.5425353646278381, "lr": 8e-05, "grad_norm": 0.23185106030492367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105329461175, "tokens_per_second": 152153.61191741884, "data_wait_s": 1.4942997950129211e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 719700, "loss": 0.568727433681488, "lr": 8e-05, "grad_norm": 0.2325424623813383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105344170815, "tokens_per_second": 152920.7540962515, "data_wait_s": 1.1985001037828624e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 719800, "loss": 0.5495955348014832, "lr": 8e-05, "grad_norm": 0.21961186793742574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105358880287, "tokens_per_second": 153857.88243120036, "data_wait_s": 1.657800021348521e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c32193da3496d6c9a21b6cf5e5710bed355bed91f09e443144f7e07d25cfdc5
-size 739809
+oid sha256:36b388b9320912b16dadaf91353057ed353431206881fbd24c000ea68951fd64
+size 741927