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": 76778.07228473613,
-      "required_pages_per_second_upper_estimate": 0.005311685190329981
+      "required_token_positions_per_second": 76622.99527260485,
+      "required_pages_per_second_upper_estimate": 0.005300956602802496
     },
     {
       "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": 9597.259035592017,
-      "required_pages_per_second_upper_estimate": 0.004018171944643929
+      "required_token_positions_per_second": 9577.874409075606,
+      "required_pages_per_second_upper_estimate": 0.00401005600631854
     },
     {
       "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": 19194.518071184033,
-      "required_pages_per_second_upper_estimate": 0.003997711920746059
+      "required_token_positions_per_second": 19155.748818151213,
+      "required_pages_per_second_upper_estimate": 0.00398963730775328
     },
     {
       "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": 28791.777106776048,
-      "required_pages_per_second_upper_estimate": 0.0017260230750530932
+      "required_token_positions_per_second": 28733.623227226817,
+      "required_pages_per_second_upper_estimate": 0.0017225368387699495
     },
     {
       "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": 28791.777106776048,
-      "required_pages_per_second_upper_estimate": 0.0022384340284082463
+      "required_token_positions_per_second": 28733.623227226817,
+      "required_pages_per_second_upper_estimate": 0.002233912820065177
     },
     {
       "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": 28791.777106776048,
-      "required_pages_per_second_upper_estimate": 0.36931000252403184
+      "required_token_positions_per_second": 28733.623227226817,
+      "required_pages_per_second_upper_estimate": 0.36856406699794536
     }
   ],
   "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": 0.049486143765623325,
-    "producer_p95_seconds": 0.13868942199997036,
-    "gpu_update_seconds": 0.9602741747223771,
-    "producer_batches_per_second": 20.207676814265586,
-    "gpu_batches_per_second": 1.0413692529939254,
-    "producer_headroom": 19.4049101758754,
+    "producer_batch_seconds": 0.057490175468707605,
+    "producer_p95_seconds": 0.15386159399986354,
+    "gpu_update_seconds": 0.9622176702658882,
+    "producer_batches_per_second": 17.39427635847639,
+    "gpu_batches_per_second": 1.0392658864014328,
+    "producer_headroom": 16.73708007361417,
     "sustainable": true,
-    "buffer_seconds": 60.497273007509754,
+    "buffer_seconds": 60.61971322675096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 802300, "loss": 0.524161159992218, "lr": 3e-05, "grad_norm": 0.26103738770913526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117433410377, "tokens_per_second": 151683.78917434896, "data_wait_s": 1.4295000255515333e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802400, "loss": 0.533500075340271, "lr": 3e-05, "grad_norm": 0.2687727625425812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117448067975, "tokens_per_second": 152703.83010156357, "data_wait_s": 2.6748999971459853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802500, "loss": 0.5279154181480408, "lr": 3e-05, "grad_norm": 0.26252601488238386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117462572395, "tokens_per_second": 150996.08411553875, "data_wait_s": 1.1963999895669986e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 802600, "loss": 0.5096468329429626, "lr": 3e-05, "grad_norm": 0.279247773655295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117477393048, "tokens_per_second": 154675.89587132793, "data_wait_s": 1.648699981160462e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29624fe5d6f74e125d0c88769582753e097be2d2d1c0bcd80ac182093376a58f
-size 53115
+oid sha256:2c965dc4ebf2a4342db666860727c246c442201bf04cfd46999885ce25733a91
+size 55233