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": 77282.53438266298,
-      "required_pages_per_second_upper_estimate": 0.005346585048777893
+      "required_token_positions_per_second": 77127.53483305838,
+      "required_pages_per_second_upper_estimate": 0.005335861820287742
     },
     {
       "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": 9660.316797832873,
-      "required_pages_per_second_upper_estimate": 0.004044572912898362
+      "required_token_positions_per_second": 9640.941854132298,
+      "required_pages_per_second_upper_estimate": 0.004036461028565763
     },
     {
       "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": 19320.633595665746,
-      "required_pages_per_second_upper_estimate": 0.004023978458605563
+      "required_token_positions_per_second": 19281.883708264595,
+      "required_pages_per_second_upper_estimate": 0.004015907878963155
     },
     {
       "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": 28980.950393498617,
-      "required_pages_per_second_upper_estimate": 0.0017373637247412273
+      "required_token_positions_per_second": 28922.82556239689,
+      "required_pages_per_second_upper_estimate": 0.001733879229868157
     },
     {
       "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": 28980.950393498617,
-      "required_pages_per_second_upper_estimate": 0.0022531414193655754
+      "required_token_positions_per_second": 28922.82556239689,
+      "required_pages_per_second_upper_estimate": 0.0022486224694115288
     },
     {
       "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": 28980.950393498617,
-      "required_pages_per_second_upper_estimate": 0.3717365143276589
+      "required_token_positions_per_second": 28922.82556239689,
+      "required_pages_per_second_upper_estimate": 0.3709909514038672
     }
   ],
   "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": 0.057878406593772525,
-    "producer_p95_seconds": 0.1468161810003039,
-    "gpu_update_seconds": 0.9540059806390048,
-    "producer_batches_per_second": 17.277600729727,
-    "gpu_batches_per_second": 1.048211458098185,
-    "producer_headroom": 16.48293442725239,
+    "producer_batch_seconds": 0.04646487638282082,
+    "producer_p95_seconds": 0.13585080800021387,
+    "gpu_update_seconds": 0.955923201222279,
+    "producer_batches_per_second": 21.52163263625348,
+    "gpu_batches_per_second": 1.0461091421584523,
+    "producer_headroom": 20.5730279651773,
     "sustainable": true,
-    "buffer_seconds": 60.1023767802573,
+    "buffer_seconds": 60.22316167700358,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 761700, "loss": 0.5007930397987366, "lr": 3e-05, "grad_norm": 0.23310555419277346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111490271322, "tokens_per_second": 152385.58246989583, "data_wait_s": 1.2020999747619499e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761800, "loss": 0.5855912566184998, "lr": 3e-05, "grad_norm": 0.27354794981320624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111505032369, "tokens_per_second": 154600.75998703312, "data_wait_s": 1.6495000181748765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761900, "loss": 0.5362855195999146, "lr": 3e-05, "grad_norm": 0.24191452364117072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111519760426, "tokens_per_second": 154197.6498428081, "data_wait_s": 9.1489996520977e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 762000, "loss": 0.5574730038642883, "lr": 3e-05, "grad_norm": 0.24691418273071314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111534488578, "tokens_per_second": 153724.32919720176, "data_wait_s": 1.8849000298359897e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f074399ef8d78431a1c04b1b123afdea05d5678d95bffb89f427eaf092e5a2d
-size 40407
+oid sha256:8c05ee9b2063edb3dd32858908c22e0299167e3f8c4880ea95def7b3a9e9d17f
+size 42525