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": 75355.42363639647,
-      "required_pages_per_second_upper_estimate": 0.00521326305583819
+      "required_token_positions_per_second": 75787.6820377265,
+      "required_pages_per_second_upper_estimate": 0.005243167695019876
     },
     {
       "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": 9419.42795454956,
-      "required_pages_per_second_upper_estimate": 0.003943717784546704
+      "required_token_positions_per_second": 9473.460254715812,
+      "required_pages_per_second_upper_estimate": 0.003966339980303581
     },
     {
       "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": 18838.85590909912,
-      "required_pages_per_second_upper_estimate": 0.0039236368718257265
+      "required_token_positions_per_second": 18946.920509431624,
+      "required_pages_per_second_upper_estimate": 0.003946143878219832
     },
     {
       "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": 28258.283863648674,
-      "required_pages_per_second_upper_estimate": 0.0016940409697246228
+      "required_token_positions_per_second": 28420.380764147438,
+      "required_pages_per_second_upper_estimate": 0.0017037584314018876
     },
     {
       "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": 28258.283863648674,
-      "required_pages_per_second_upper_estimate": 0.0021969572753438747
+      "required_token_positions_per_second": 28420.380764147438,
+      "required_pages_per_second_upper_estimate": 0.002209559596368739
     },
     {
       "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": 28258.283863648674,
-      "required_pages_per_second_upper_estimate": 0.3624669240216092
+      "required_token_positions_per_second": 28420.380764147438,
+      "required_pages_per_second_upper_estimate": 0.36454612901511574
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3242878789609165,
-    "producer_p95_seconds": 0.23220220300026995,
-    "gpu_update_seconds": 0.9784033642455641,
-    "producer_batches_per_second": 3.0836798563184074,
-    "gpu_batches_per_second": 1.022073345762756,
-    "producer_headroom": 3.0170827456782074,
+    "producer_batch_seconds": 0.07216260640618799,
+    "producer_p95_seconds": 0.21163548399999854,
+    "gpu_update_seconds": 0.9728229973216332,
+    "producer_batches_per_second": 13.857592592640186,
+    "gpu_batches_per_second": 1.0279362255551012,
+    "producer_headroom": 13.48098476163429,
     "sustainable": true,
-    "buffer_seconds": 60.66100858322497,
+    "buffer_seconds": 60.31502583394126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371711.91680161195,
-    "upload_bytes_per_second": 3998177.236528712,
+    "generation_bytes_per_second": 370914.2270079003,
+    "upload_bytes_per_second": 4037284.2097015153,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.76072433506269,
+    "estimated_full_shards_per_hour": 79.58955867460018,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 6089737,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.5443801999999778,
+      "bytes": 7496481,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -64,3 +64,4 @@
 {"step": 806400, "loss": 0.5340363383293152, "lr": 3e-05, "grad_norm": 0.2616473244955081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118033425331, "tokens_per_second": 149681.91922811183, "data_wait_s": 2.4187999770219903e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806500, "loss": 0.5248042941093445, "lr": 3e-05, "grad_norm": 0.2763862277188099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118047954900, "tokens_per_second": 149281.52852920647, "data_wait_s": 1.666000025579706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806600, "loss": 0.5344794392585754, "lr": 3e-05, "grad_norm": 0.2699462368538555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118062591116, "tokens_per_second": 150935.447145856, "data_wait_s": 1.5345999599958304e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 806700, "loss": 0.536081075668335, "lr": 3e-05, "grad_norm": 0.2757851306726722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118077248182, "tokens_per_second": 151394.7045601352, "data_wait_s": 1.7426999875169713e-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:af990e0ce682e75adcb8e653cc36c7b4d2c01ec16c4988952b508a3e86d74308
-size 140736
+oid sha256:b47c53752959d51712ff2fa28ec63742fedb42f680326d4ecf51b8162cc2607f
+size 142854