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": 76748.43504402942,
-      "required_pages_per_second_upper_estimate": 0.005309634817250018
+      "required_token_positions_per_second": 77741.91624564744,
+      "required_pages_per_second_upper_estimate": 0.005378366151971934
     },
     {
       "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": 9593.554380503678,
-      "required_pages_per_second_upper_estimate": 0.004016620882920437
+      "required_token_positions_per_second": 9717.73953070593,
+      "required_pages_per_second_upper_estimate": 0.004068614612029292
     },
     {
       "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": 19187.108761007355,
-      "required_pages_per_second_upper_estimate": 0.003996168756832907
+      "required_token_positions_per_second": 19435.47906141186,
+      "required_pages_per_second_upper_estimate": 0.004047897740441878
     },
     {
       "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": 28780.66314151103,
-      "required_pages_per_second_upper_estimate": 0.0017253568098054978
+      "required_token_positions_per_second": 29153.21859211779,
+      "required_pages_per_second_upper_estimate": 0.0017476909402101381
     },
     {
       "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": 28780.66314151103,
-      "required_pages_per_second_upper_estimate": 0.0022375699664940584
+      "required_token_positions_per_second": 29153.21859211779,
+      "required_pages_per_second_upper_estimate": 0.0022665345140804897
     },
     {
       "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": 28780.66314151103,
-      "required_pages_per_second_upper_estimate": 0.3691674445108584
+      "required_token_positions_per_second": 29153.21859211779,
+      "required_pages_per_second_upper_estimate": 0.37394618581236505
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.061926396843773546,
-    "producer_p95_seconds": 0.21567140799743356,
-    "gpu_update_seconds": 0.9606449950113427,
-    "producer_batches_per_second": 16.14820255928625,
-    "gpu_batches_per_second": 1.0409672721900691,
-    "producer_headroom": 15.512689967007692,
+    "producer_batch_seconds": 0.10565764236736186,
+    "producer_p95_seconds": 0.22741282800052431,
+    "gpu_update_seconds": 0.9483686994160995,
+    "producer_batches_per_second": 9.464530701178173,
+    "gpu_batches_per_second": 1.05444222338389,
+    "producer_headroom": 8.975864671660089,
     "sustainable": true,
-    "buffer_seconds": 60.52063468571459,
+    "buffer_seconds": 60.695596762630366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272870.12847277854,
-    "upload_bytes_per_second": 5599824.325670854,
+    "generation_bytes_per_second": 285561.0979242615,
+    "upload_bytes_per_second": 5474101.035083172,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.5515774787666,
+    "estimated_full_shards_per_hour": 61.27476409240612,
     "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": 1.1637235170001077,
-      "bytes": 10485207,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.0128666689997772,
       "bytes": 6455974,
@@ -562,6 +556,12 @@
       "bytes": 15692090,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.250901432998944,
+      "bytes": 1087271,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -185,3 +185,4 @@
 {"step": 543500, "loss": 0.6088278293609619, "lr": 0.0003, "grad_norm": 0.16523491964644105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79551005916, "tokens_per_second": 113077.61009661564, "data_wait_s": 8.88299837242812e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 543600, "loss": 0.5791146159172058, "lr": 0.0003, "grad_norm": 0.1897782516420057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79565626298, "tokens_per_second": 152548.50523329302, "data_wait_s": 1.2267002603039145e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 543700, "loss": 0.5663001537322998, "lr": 0.0003, "grad_norm": 0.14818047548669092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79580149359, "tokens_per_second": 152315.03329122846, "data_wait_s": 1.9553001038730145e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 543800, "loss": 0.5979887247085571, "lr": 0.0003, "grad_norm": 0.17746552094419787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79594737921, "tokens_per_second": 152411.5152703311, "data_wait_s": 9.181003406411037e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ddb22bc84bb40bfb86e625ec1bea5fb7012fc12be6b55343025c3bddc7905f6a
-size 398580
+oid sha256:ecaa955eaeccba42f30f74768dff7c5c9c80a143ded24eef1ede3cd75786667f
+size 400698