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": 76014.26962170882,
-      "required_pages_per_second_upper_estimate": 0.005258843549835403
+      "required_token_positions_per_second": 76286.28091363392,
+      "required_pages_per_second_upper_estimate": 0.005277661922163934
     },
     {
       "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": 9501.783702713603,
-      "required_pages_per_second_upper_estimate": 0.0039781984165194045
+      "required_token_positions_per_second": 9535.78511420424,
+      "required_pages_per_second_upper_estimate": 0.003992434097480322
     },
     {
       "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": 19003.567405427206,
-      "required_pages_per_second_upper_estimate": 0.0039579419327766055
+      "required_token_positions_per_second": 19071.57022840848,
+      "required_pages_per_second_upper_estimate": 0.003972105127448565
     },
     {
       "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": 28505.351108140807,
-      "required_pages_per_second_upper_estimate": 0.0017088522737820871
+      "required_token_positions_per_second": 28607.355342612722,
+      "required_pages_per_second_upper_estimate": 0.0017149672718872308
     },
     {
       "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": 28505.351108140807,
-      "required_pages_per_second_upper_estimate": 0.0022161656668691788
+      "required_token_positions_per_second": 28607.355342612722,
+      "required_pages_per_second_upper_estimate": 0.0022240960474301597
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28505.351108140807,
-      "required_pages_per_second_upper_estimate": 0.36563603735381545
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28607.355342612722,
+      "required_pages_per_second_upper_estimate": 0.3669444381500073
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16298761008599172,
-    "producer_p95_seconds": 0.26388264899651404,
-    "gpu_update_seconds": 0.9699231521517391,
-    "producer_batches_per_second": 6.135435690310468,
-    "gpu_batches_per_second": 1.0310095163534725,
-    "producer_headroom": 5.95090112457021,
+    "producer_batch_seconds": 0.12003513783585618,
+    "producer_p95_seconds": 0.258319336000568,
+    "gpu_update_seconds": 0.9664647314956901,
+    "producer_batches_per_second": 8.330893920140824,
+    "gpu_batches_per_second": 1.0346989056211198,
+    "producer_headroom": 8.051515155647978,
     "sustainable": true,
-    "buffer_seconds": 60.13523543340782,
+    "buffer_seconds": 60.88727808422848,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 275171.36867459066,
-    "upload_bytes_per_second": 4000991.639235824,
+    "generation_bytes_per_second": 288124.76647762296,
+    "upload_bytes_per_second": 3945111.9080191674,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.0453700559453,
+    "estimated_full_shards_per_hour": 61.824867684807934,
     "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": 1.6420892379992438,
-      "bytes": 7607765,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.5253431410001212,
       "bytes": 5688723,
@@ -562,6 +556,12 @@
       "bytes": 16328104,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.913415155999246,
+      "bytes": 2704493,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -218,3 +218,4 @@
 {"step": 821800, "loss": 0.4260617196559906, "lr": 3e-05, "grad_norm": 0.27389021874056513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120288473389, "tokens_per_second": 129560.19725595282, "data_wait_s": 1.0712999937823042e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821900, "loss": 0.5479665994644165, "lr": 3e-05, "grad_norm": 0.29453559169036236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120303185922, "tokens_per_second": 151038.70904621444, "data_wait_s": 1.0588999430183321e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822000, "loss": 0.5329545736312866, "lr": 3e-05, "grad_norm": 0.30049129939144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120317816022, "tokens_per_second": 151462.89790822644, "data_wait_s": 1.4623001334257424e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 822100, "loss": 0.5251942873001099, "lr": 3e-05, "grad_norm": 0.286140653402896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120332322134, "tokens_per_second": 148917.0294094196, "data_wait_s": 1.5284000255633146e-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:e5be1933294ebcf6e6a1327815149acc771afd87d62ce5ac8dfc785925d4b4c5
-size 469257
+oid sha256:d5f3095da0a3fbcfb95e0f0e8a59afe3a89ac5b2e9350f00734f9a75965decce
+size 471375