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": 77354.88831734609,
-      "required_pages_per_second_upper_estimate": 0.005351590661863529
+      "required_token_positions_per_second": 76944.82551815371,
+      "required_pages_per_second_upper_estimate": 0.005323221565938618
     },
     {
       "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": 9669.361039668262,
-      "required_pages_per_second_upper_estimate": 0.004048359548089593
+      "required_token_positions_per_second": 9618.103189769214,
+      "required_pages_per_second_upper_estimate": 0.004026898956722485
     },
     {
       "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": 19338.722079336523,
-      "required_pages_per_second_upper_estimate": 0.004027745812728793
+      "required_token_positions_per_second": 19236.206379538427,
+      "required_pages_per_second_upper_estimate": 0.004006394495981658
     },
     {
       "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": 29008.083119004783,
-      "required_pages_per_second_upper_estimate": 0.001738990290206049
+      "required_token_positions_per_second": 28854.309569307643,
+      "required_pages_per_second_upper_estimate": 0.0017297718007003207
     },
     {
       "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": 29008.083119004783,
-      "required_pages_per_second_upper_estimate": 0.002255250869429433
+      "required_token_positions_per_second": 28854.309569307643,
+      "required_pages_per_second_upper_estimate": 0.0022432956523188437
     },
     {
       "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": 29008.083119004783,
-      "required_pages_per_second_upper_estimate": 0.37208454379760114
+      "required_token_positions_per_second": 28854.309569307643,
+      "required_pages_per_second_upper_estimate": 0.37011210181126003
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.050129372335941724,
-    "producer_p95_seconds": 0.1267673410000043,
-    "gpu_update_seconds": 0.9531136506529893,
-    "producer_batches_per_second": 19.94838461767495,
-    "gpu_batches_per_second": 1.0491928211445596,
-    "producer_headroom": 19.013077687582104,
+    "producer_batch_seconds": 0.04716634035938583,
+    "producer_p95_seconds": 0.12113686200018492,
+    "gpu_update_seconds": 0.9581930884046937,
+    "producer_batches_per_second": 21.201560103676897,
+    "gpu_batches_per_second": 1.043630988473222,
+    "producer_headroom": 20.315188354739906,
     "sustainable": true,
-    "buffer_seconds": 60.046159991138325,
+    "buffer_seconds": 60.36616456949571,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 70,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -16,3 +16,4 @@
 {"step": 926600, "loss": 0.5047388076782227, "lr": 3e-05, "grad_norm": 0.31476835165476447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135627853641, "tokens_per_second": 152753.400048464, "data_wait_s": 7.586000037917984e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 926700, "loss": 0.5236449241638184, "lr": 3e-05, "grad_norm": 0.334416242186587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135642531350, "tokens_per_second": 154168.46086430972, "data_wait_s": 8.987999990495155e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 926800, "loss": 0.5487688183784485, "lr": 3e-05, "grad_norm": 0.356873609202194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135657183960, "tokens_per_second": 152888.063818093, "data_wait_s": 8.877999789547175e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 926900, "loss": 0.553955078125, "lr": 3e-05, "grad_norm": 0.3240316203280456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135671791690, "tokens_per_second": 153263.09404549634, "data_wait_s": 8.334000085596927e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4dd1b96265c219a3f6261867655548723ca634576f45c5b3fd40d0143681e45a
-size 38289
+oid sha256:ad9ef968e78fe2008185fd68cf4a52672692322cff816a6c834f58da41771814
+size 40407