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": 77114.3134956461,
-      "required_pages_per_second_upper_estimate": 0.005334947137487831
+      "required_token_positions_per_second": 75903.9666235327,
+      "required_pages_per_second_upper_estimate": 0.005251212532483362
     },
     {
       "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": 9639.289186955762,
-      "required_pages_per_second_upper_estimate": 0.004035769091330564
+      "required_token_positions_per_second": 9487.995827941588,
+      "required_pages_per_second_upper_estimate": 0.003972425721275928
     },
     {
       "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": 19278.578373911525,
-      "required_pages_per_second_upper_estimate": 0.004015219464984939
+      "required_token_positions_per_second": 18975.991655883176,
+      "required_pages_per_second_upper_estimate": 0.003952198631368003
     },
     {
       "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": 28917.867560867286,
-      "required_pages_per_second_upper_estimate": 0.0017335820052468873
+      "required_token_positions_per_second": 28463.987483824763,
+      "required_pages_per_second_upper_estimate": 0.0017063725876629377
     },
     {
       "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": 28917.867560867286,
-      "required_pages_per_second_upper_estimate": 0.002248237006600546
+      "required_token_positions_per_second": 28463.987483824763,
+      "required_pages_per_second_upper_estimate": 0.002212949827017962
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28917.867560867286,
-      "required_pages_per_second_upper_estimate": 0.37092735548373273
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28463.987483824763,
+      "required_pages_per_second_upper_estimate": 0.3651054691938888
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1097967353749425,
-    "producer_p95_seconds": 0.29981810100071016,
-    "gpu_update_seconds": 0.9560870953504982,
-    "producer_batches_per_second": 9.107738919423438,
-    "gpu_batches_per_second": 1.045929816292943,
-    "producer_headroom": 8.70779164868224,
+    "producer_batch_seconds": 0.2596490188124676,
+    "producer_p95_seconds": 0.270526478998363,
+    "gpu_update_seconds": 0.9713326362201197,
+    "producer_batches_per_second": 3.851352893893481,
+    "gpu_batches_per_second": 1.0295134361915785,
+    "producer_headroom": 3.7409447594395417,
     "sustainable": true,
-    "buffer_seconds": 60.233487007081386,
+    "buffer_seconds": 60.22262344564742,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416562.57778706355,
-    "upload_bytes_per_second": 5702473.091301636,
+    "generation_bytes_per_second": 404714.16084463504,
+    "upload_bytes_per_second": 5697241.510768052,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.3846321125882,
+    "estimated_full_shards_per_hour": 86.8422376537732,
     "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": 1.029109121998772,
-      "bytes": 7643162,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.8460896450014843,
       "bytes": 884554,
@@ -562,6 +556,12 @@
       "bytes": 5817766,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0792979020006896,
+      "bytes": 7541421,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -296,3 +296,4 @@
 {"step": 586700, "loss": 0.5536483526229858, "lr": 0.0003, "grad_norm": 0.14367783967544867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85875449788, "tokens_per_second": 152037.35700774894, "data_wait_s": 8.17299951449968e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 586800, "loss": 0.5427125692367554, "lr": 0.0003, "grad_norm": 0.13986188715868894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85890049418, "tokens_per_second": 152087.50209607932, "data_wait_s": 1.151100150309503e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 586900, "loss": 0.5674201250076294, "lr": 0.0003, "grad_norm": 0.151298944953679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85904721763, "tokens_per_second": 152233.33595929542, "data_wait_s": 6.4159976318478584e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 587000, "loss": 0.5778745412826538, "lr": 0.0003, "grad_norm": 0.16915777713992394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85919437460, "tokens_per_second": 152308.90628709272, "data_wait_s": 1.4666999049950391e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dfc438a592514e64d9109be02e75d66589e71593d319c9292e5c8b234d30a8cb
-size 636027
+oid sha256:70ad8ce697e126dc45942f704c3ddc28218c7f4602225856079442b40e35ef15
+size 638145