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": 76697.53429879525,
-      "required_pages_per_second_upper_estimate": 0.005306113385588717
+      "required_token_positions_per_second": 77364.5153886265,
+      "required_pages_per_second_upper_estimate": 0.005352256684992591
     },
     {
       "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": 9587.191787349406,
-      "required_pages_per_second_upper_estimate": 0.004013956998032797
+      "required_token_positions_per_second": 9670.564423578313,
+      "required_pages_per_second_upper_estimate": 0.004048863379803218
     },
     {
       "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": 19174.38357469881,
-      "required_pages_per_second_upper_estimate": 0.0039935184361105675
+      "required_token_positions_per_second": 19341.128847156626,
+      "required_pages_per_second_upper_estimate": 0.0040282470789949855
     },
     {
       "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": 28761.575362048217,
-      "required_pages_per_second_upper_estimate": 0.001724212526572054
+      "required_token_positions_per_second": 29011.693270734937,
+      "required_pages_per_second_upper_estimate": 0.0017392067132898893
     },
     {
       "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": 28761.575362048217,
-      "required_pages_per_second_upper_estimate": 0.0022360859755990933
+      "required_token_positions_per_second": 29011.693270734937,
+      "required_pages_per_second_upper_estimate": 0.002255531542847073
     },
     {
       "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": 28761.575362048217,
-      "required_pages_per_second_upper_estimate": 0.36892260697077023
+      "required_token_positions_per_second": 29011.693270734937,
+      "required_pages_per_second_upper_estimate": 0.37213085094771664
     }
   ],
   "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": 0.07329332001577882,
-    "producer_p95_seconds": 0.2700719599997683,
-    "gpu_update_seconds": 0.9612825324054532,
-    "producer_batches_per_second": 13.643808191315618,
-    "gpu_batches_per_second": 1.040276886648156,
-    "producer_headroom": 13.115554489802145,
+    "producer_batch_seconds": 0.41876649125046583,
+    "producer_p95_seconds": 0.28375503700226545,
+    "gpu_update_seconds": 0.9529950472725237,
+    "producer_batches_per_second": 2.3879656584124738,
+    "gpu_batches_per_second": 1.0493233966556328,
+    "producer_headroom": 2.2757194455239587,
     "sustainable": true,
-    "buffer_seconds": 60.56079954154355,
+    "buffer_seconds": 60.038687978168994,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379776.6634626195,
-    "upload_bytes_per_second": 4777748.649648323,
+    "generation_bytes_per_second": 346496.04825219046,
+    "upload_bytes_per_second": 4717095.948689322,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.49123123082101,
+    "estimated_full_shards_per_hour": 74.34998593973432,
     "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": 0.9455101860003197,
-      "bytes": 6297280,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.347191354001552,
       "bytes": 5982017,
@@ -562,6 +556,12 @@
       "bytes": 11893185,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.8795082219985488,
+      "bytes": 997738,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -304,3 +304,4 @@
 {"step": 715400, "loss": 0.5608413219451904, "lr": 8e-05, "grad_norm": 0.21070986636697486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104714675976, "tokens_per_second": 152025.3985217592, "data_wait_s": 1.5050998626975343e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715500, "loss": 0.5825920104980469, "lr": 8e-05, "grad_norm": 0.21111483230852543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104729406845, "tokens_per_second": 153312.53763521224, "data_wait_s": 1.0243998985970393e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715600, "loss": 0.5703750848770142, "lr": 8e-05, "grad_norm": 0.22452769669210665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104744145932, "tokens_per_second": 153219.24942496963, "data_wait_s": 1.5197001630440354e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 715700, "loss": 0.6353872418403625, "lr": 8e-05, "grad_norm": 0.27269471104659276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104758954617, "tokens_per_second": 153607.3803993817, "data_wait_s": 1.4653000107500702e-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:d9fce86006ced37f2d91f5a6d398a9c18fa0bd6f1178c4a56425541a14a7cf78
-size 652971
+oid sha256:684d465714c04cacd15b5f8bd1d7658c7c7c51d56adcf8898efb64920dc62fbb
+size 655089