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": 77616.0525016989,
-      "required_pages_per_second_upper_estimate": 0.005369658606121452
+      "required_token_positions_per_second": 75991.0801460436,
+      "required_pages_per_second_upper_estimate": 0.005257239248101887
     },
     {
       "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": 9702.006562712362,
-      "required_pages_per_second_upper_estimate": 0.004062027546872116
+      "required_token_positions_per_second": 9498.88501825545,
+      "required_pages_per_second_upper_estimate": 0.003976984797868191
     },
     {
       "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": 19404.013125424724,
-      "required_pages_per_second_upper_estimate": 0.004041344215788291
+      "required_token_positions_per_second": 18997.7700365109,
+      "required_pages_per_second_upper_estimate": 0.003956734493718238
     },
     {
       "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": 29106.019688137083,
-      "required_pages_per_second_upper_estimate": 0.001744861437985048
+      "required_token_positions_per_second": 28496.65505476635,
+      "required_pages_per_second_upper_estimate": 0.0017083309586603931
     },
     {
       "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": 29106.019688137083,
-      "required_pages_per_second_upper_estimate": 0.0022628650069020277
+      "required_token_positions_per_second": 28496.65505476635,
+      "required_pages_per_second_upper_estimate": 0.0022154895869692113
     },
     {
       "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": 29106.019688137083,
-      "required_pages_per_second_upper_estimate": 0.3733407689503352
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28496.65505476635,
+      "required_pages_per_second_upper_estimate": 0.3655244937182226
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12440547867925034,
-    "producer_p95_seconds": 0.3491304909985047,
-    "gpu_update_seconds": 0.9499065930773304,
-    "producer_batches_per_second": 8.038231198629603,
-    "gpu_batches_per_second": 1.0527350870998655,
-    "producer_headroom": 7.635568812258152,
+    "producer_batch_seconds": 0.2681699248831251,
+    "producer_p95_seconds": 0.3320305599991116,
+    "gpu_update_seconds": 0.9702191343813735,
+    "producer_batches_per_second": 3.728978931682306,
+    "gpu_batches_per_second": 1.0306949889600097,
+    "producer_headroom": 3.617926711223186,
     "sustainable": true,
-    "buffer_seconds": 60.794021956949145,
+    "buffer_seconds": 60.15358633164516,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294381.5781204952,
-    "upload_bytes_per_second": 4981828.230443674,
+    "generation_bytes_per_second": 293934.8236481574,
+    "upload_bytes_per_second": 4903113.124838578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.16743381224767,
+    "estimated_full_shards_per_hour": 63.07157070239584,
     "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": 0.9838692140037892,
-      "bytes": 6891211,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.3423998879952705,
       "bytes": 11990725,
@@ -562,6 +556,12 @@
       "bytes": 5863415,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 2.5138282220068504,
+      "bytes": 7584714,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -423,3 +423,4 @@
 {"step": 927300, "loss": 0.5017232894897461, "lr": 3e-05, "grad_norm": 0.3300251191732454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135730323364, "tokens_per_second": 151996.25186029857, "data_wait_s": 1.6120000509545207e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927400, "loss": 0.5002279281616211, "lr": 3e-05, "grad_norm": 0.32860060915400224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135745010409, "tokens_per_second": 153240.5829542607, "data_wait_s": 1.6832003893796355e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927500, "loss": 0.523468554019928, "lr": 3e-05, "grad_norm": 0.34131293320080286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135759655387, "tokens_per_second": 152516.0676678596, "data_wait_s": 8.820999937597662e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 927600, "loss": 0.5498864650726318, "lr": 3e-05, "grad_norm": 0.3313419076080202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135774307827, "tokens_per_second": 153132.47373963153, "data_wait_s": 1.359300222247839e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c7d3897c403aeebdec859e5f59a524bbd2b68453168896aa2031d56e0713801
-size 906579
+oid sha256:5a669ce431b9d8137b7c9d73f4222fb1ccb0bf6d9c551616b00102da5a56d716
+size 908697