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": 77443.62378682963,
-      "required_pages_per_second_upper_estimate": 0.005357729587537056
+      "required_token_positions_per_second": 76978.51777852977,
+      "required_pages_per_second_upper_estimate": 0.005325552474688249
     },
     {
       "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": 9680.452973353704,
-      "required_pages_per_second_upper_estimate": 0.004053003509097773
+      "required_token_positions_per_second": 9622.314722316221,
+      "required_pages_per_second_upper_estimate": 0.004028662237453117
     },
     {
       "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": 19360.905946707408,
-      "required_pages_per_second_upper_estimate": 0.004032366127274225
+      "required_token_positions_per_second": 19244.629444632443,
+      "required_pages_per_second_upper_estimate": 0.004008148798309577
     },
     {
       "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": 29041.358920061113,
-      "required_pages_per_second_upper_estimate": 0.0017409851236701698
+      "required_token_positions_per_second": 28866.944166948666,
+      "required_pages_per_second_upper_estimate": 0.0017305292255369864
     },
     {
       "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": 29041.358920061113,
-      "required_pages_per_second_upper_estimate": 0.0022578379165967824
+      "required_token_positions_per_second": 28866.944166948666,
+      "required_pages_per_second_upper_estimate": 0.0022442779367117115
     },
     {
       "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": 29041.358920061113,
-      "required_pages_per_second_upper_estimate": 0.37251137004478024
+      "required_token_positions_per_second": 28866.944166948666,
+      "required_pages_per_second_upper_estimate": 0.3702741648638251
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.19820812310206293,
-    "producer_p95_seconds": 0.2819794900060515,
-    "gpu_update_seconds": 0.9520215660742166,
-    "producer_batches_per_second": 5.04520190368319,
-    "gpu_batches_per_second": 1.0503963729767474,
-    "producer_headroom": 4.80314101750509,
+    "producer_batch_seconds": 0.09709957758627752,
+    "producer_p95_seconds": 0.27649424700211966,
+    "gpu_update_seconds": 0.9577737026857072,
+    "producer_batches_per_second": 10.298705976464761,
+    "gpu_batches_per_second": 1.044087969001326,
+    "producer_headroom": 9.863829755950075,
     "sustainable": true,
-    "buffer_seconds": 60.929380228749864,
+    "buffer_seconds": 60.33974326919955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389873.88519100065,
-    "upload_bytes_per_second": 4908943.645002797,
+    "generation_bytes_per_second": 391567.86624778516,
+    "upload_bytes_per_second": 4990016.6015165225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.65785996243967,
+    "estimated_full_shards_per_hour": 84.02134886336485,
     "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": 1.2893916930006526,
-      "bytes": 6001693,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.5835260300009395,
       "bytes": 761598,
@@ -562,6 +556,12 @@
       "bytes": 12201541,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.149907068000175,
+      "bytes": 11779236,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -310,3 +310,4 @@
 {"step": 716000, "loss": 0.5255452990531921, "lr": 8e-05, "grad_norm": 0.2099402970839708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104802603951, "tokens_per_second": 152256.716971786, "data_wait_s": 1.0746000043582171e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 716100, "loss": 0.542158305644989, "lr": 8e-05, "grad_norm": 0.23822071220199037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104817278984, "tokens_per_second": 152792.09080238626, "data_wait_s": 1.608499587746337e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 716200, "loss": 0.5775470733642578, "lr": 8e-05, "grad_norm": 0.2149234666184996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104831966725, "tokens_per_second": 152899.9567920593, "data_wait_s": 1.5219004126265645e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 716300, "loss": 0.527855396270752, "lr": 8e-05, "grad_norm": 0.21463655181509075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104846597691, "tokens_per_second": 151524.48098634294, "data_wait_s": 1.617900124983862e-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:a1f3022006f8e4f18cc040e1f33e6acfed5332bbb1e9754592e3eb76d41166c7
-size 665679
+oid sha256:f81c7ef39920d224907e34245233ee525e3b0856bc71fc6e11f1927385e89f8c
+size 667797