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": 76184.14699571164,
-      "required_pages_per_second_upper_estimate": 0.0052705960607387315
+      "required_token_positions_per_second": 75698.42086348265,
+      "required_pages_per_second_upper_estimate": 0.0052369923998712275
     },
     {
       "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": 9523.018374463954,
-      "required_pages_per_second_upper_estimate": 0.0039870889301509066
+      "required_token_positions_per_second": 9462.302607935331,
+      "required_pages_per_second_upper_estimate": 0.003961668506594754
     },
     {
       "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": 19046.03674892791,
-      "required_pages_per_second_upper_estimate": 0.0039667871770358975
+      "required_token_positions_per_second": 18924.605215870663,
+      "required_pages_per_second_upper_estimate": 0.0039414961910649495
     },
     {
       "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": 28569.055123391867,
-      "required_pages_per_second_upper_estimate": 0.0017126712322259891
+      "required_token_positions_per_second": 28386.907823805992,
+      "required_pages_per_second_upper_estimate": 0.0017017517797386385
     },
     {
       "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": 28569.055123391867,
-      "required_pages_per_second_upper_estimate": 0.0022211183738505987
+      "required_token_positions_per_second": 28386.907823805992,
+      "required_pages_per_second_upper_estimate": 0.0022069572224890964
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28569.055123391867,
-      "required_pages_per_second_upper_estimate": 0.3664531640614136
+      "required_token_positions_per_second": 28386.907823805992,
+      "required_pages_per_second_upper_estimate": 0.36411677407685883
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3121420601799798,
-    "producer_p95_seconds": 0.3306264970015036,
-    "gpu_update_seconds": 0.9677603925151266,
-    "producer_batches_per_second": 3.203669506837381,
-    "gpu_batches_per_second": 1.033313625701384,
-    "producer_headroom": 3.100384459425686,
+    "producer_batch_seconds": 0.13373144945319382,
+    "producer_p95_seconds": 0.3377000010004849,
+    "gpu_update_seconds": 0.9739701193102009,
+    "producer_batches_per_second": 7.477672634887587,
+    "gpu_batches_per_second": 1.0267255433957607,
+    "producer_headroom": 7.283029708364088,
     "sustainable": true,
-    "buffer_seconds": 60.001144335937845,
+    "buffer_seconds": 60.386147397232456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491888.49617113854,
-    "upload_bytes_per_second": 4761881.979657622,
+    "generation_bytes_per_second": 510990.6168915106,
+    "upload_bytes_per_second": 4820846.4135703975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.5478207001745,
+    "estimated_full_shards_per_hour": 109.64669113215436,
     "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": 1.4130519919999642,
-      "bytes": 6156286,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9332762750018446,
       "bytes": 1158586,
@@ -562,6 +556,12 @@
       "bytes": 6331179,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.503990763994807,
+      "bytes": 11868685,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -334,3 +334,4 @@
 {"step": 958400, "loss": 0.5396652221679688, "lr": 3e-05, "grad_norm": 0.33262629903828717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140283856656, "tokens_per_second": 152832.01059655382, "data_wait_s": 9.65100116445683e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 958500, "loss": 0.5348881483078003, "lr": 3e-05, "grad_norm": 0.3582072173167426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140298567273, "tokens_per_second": 152959.63421509392, "data_wait_s": 1.603599957888946e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 958600, "loss": 0.5070256590843201, "lr": 3e-05, "grad_norm": 0.3345787106480922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140313293116, "tokens_per_second": 152514.82561405018, "data_wait_s": 8.761999197304249e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 958700, "loss": 0.506074845790863, "lr": 3e-05, "grad_norm": 0.3511785905832599, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140327799300, "tokens_per_second": 151487.13559697891, "data_wait_s": 1.4682002074550837e-05, "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 sha256:011aef023bd6723def402ecdbd1826aef9d88012ed03c6bebcfcf6ce51889b2b
-size 716511
+oid sha256:8d917ec5668353148304a444ac0f291d266f2d9cf5f01e143b227bb3cdd9df67
+size 718629