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": 77480.86716235138,
-      "required_pages_per_second_upper_estimate": 0.005360306170672188
+      "required_token_positions_per_second": 77443.87077223754,
+      "required_pages_per_second_upper_estimate": 0.005357746674560661
     },
     {
       "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": 9685.108395293923,
-      "required_pages_per_second_upper_estimate": 0.004054952637047878
+      "required_token_positions_per_second": 9680.483846529693,
+      "required_pages_per_second_upper_estimate": 0.004053016435051854
     },
     {
       "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": 19370.216790587845,
-      "required_pages_per_second_upper_estimate": 0.0040343053305110545
+      "required_token_positions_per_second": 19360.967693059385,
+      "required_pages_per_second_upper_estimate": 0.004032378987410981
     },
     {
       "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": 29055.325185881768,
-      "required_pages_per_second_upper_estimate": 0.001741822379980865
+      "required_token_positions_per_second": 29041.451539589078,
+      "required_pages_per_second_upper_estimate": 0.0017409906760694495
     },
     {
       "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": 29055.325185881768,
-      "required_pages_per_second_upper_estimate": 0.002258923732333227
+      "required_token_positions_per_second": 29041.451539589078,
+      "required_pages_per_second_upper_estimate": 0.0022578451173576914
     },
     {
       "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": 29055.325185881768,
-      "required_pages_per_second_upper_estimate": 0.3726905143069197
+      "required_token_positions_per_second": 29041.451539589078,
+      "required_pages_per_second_upper_estimate": 0.37251255806863787
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04876409598432119,
-    "producer_p95_seconds": 0.1429294250001476,
-    "gpu_update_seconds": 0.9515639499169811,
-    "producer_batches_per_second": 20.506890978180415,
-    "gpu_batches_per_second": 1.0509015185865802,
-    "producer_headroom": 19.51361817971426,
+    "producer_batch_seconds": 0.05888907927338494,
+    "producer_p95_seconds": 0.16834311700040416,
+    "gpu_update_seconds": 0.9520185298696404,
+    "producer_batches_per_second": 16.981077176595498,
+    "gpu_batches_per_second": 1.0503997229307394,
+    "producer_headroom": 16.166300129265352,
     "sustainable": true,
-    "buffer_seconds": 60.90009279468679,
+    "buffer_seconds": 60.92918591165699,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -38,3 +38,4 @@
 {"step": 688800, "loss": 0.5484572052955627, "lr": 8e-05, "grad_norm": 0.1872351369526101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100821681614, "tokens_per_second": 152381.7719375346, "data_wait_s": 1.6102000245155068e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688900, "loss": 0.5259572863578796, "lr": 8e-05, "grad_norm": 0.1889048037181038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100836258126, "tokens_per_second": 152824.50179066346, "data_wait_s": 1.6406000213464722e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689000, "loss": 0.5641706585884094, "lr": 8e-05, "grad_norm": 0.1990296048883631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100850908471, "tokens_per_second": 153449.5390437188, "data_wait_s": 1.5900999642326497e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 689100, "loss": 0.5324979424476624, "lr": 8e-05, "grad_norm": 0.20540835619071118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100865561638, "tokens_per_second": 153379.23205891647, "data_wait_s": 1.6363999748136848e-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 sha256:5fd1f4a2e618f3e93d4c19f912ee016a564003b939461e79ccaea87c73705ef1
-size 84885
+oid sha256:6649d88abbdfad777ab9a4bb3f826dee84261091104ff213d1a4d755b641b742
+size 87003