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": 77476.24468758787,
-      "required_pages_per_second_upper_estimate": 0.005359986377142429
+      "required_token_positions_per_second": 77288.03065043864,
+      "required_pages_per_second_upper_estimate": 0.005346965293335711
     },
     {
       "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": 9684.530585948483,
-      "required_pages_per_second_upper_estimate": 0.004054710720340974
+      "required_token_positions_per_second": 9661.00383130483,
+      "required_pages_per_second_upper_estimate": 0.00404486055946618
     },
     {
       "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": 19369.061171896967,
-      "required_pages_per_second_upper_estimate": 0.004034064645613459
+      "required_token_positions_per_second": 19322.00766260966,
+      "required_pages_per_second_upper_estimate": 0.004024264640513399
     },
     {
       "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": 29053.591757845446,
-      "required_pages_per_second_upper_estimate": 0.0017417184636168788
+      "required_token_positions_per_second": 28983.011493914488,
+      "required_pages_per_second_upper_estimate": 0.0017374872845640407
     },
     {
       "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": 29053.591757845446,
-      "required_pages_per_second_upper_estimate": 0.0022587889659279473
+      "required_token_positions_per_second": 28983.011493914488,
+      "required_pages_per_second_upper_estimate": 0.0022533016608570866
     },
     {
       "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": 29053.591757845446,
-      "required_pages_per_second_upper_estimate": 0.3726682797532798
+      "required_token_positions_per_second": 28983.011493914488,
+      "required_pages_per_second_upper_estimate": 0.3717629519107565
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.15527168846094241,
-    "producer_p95_seconds": 0.21671986000001198,
-    "gpu_update_seconds": 0.9516207231945465,
-    "producer_batches_per_second": 6.44032411775791,
-    "gpu_batches_per_second": 1.0508388222600349,
-    "producer_headroom": 6.128745894548062,
+    "producer_batch_seconds": 0.16258202030471125,
+    "producer_p95_seconds": 0.214224371000455,
+    "gpu_update_seconds": 0.9539381373742063,
+    "producer_batches_per_second": 6.150741626446761,
+    "gpu_batches_per_second": 1.048286006000958,
+    "producer_headroom": 5.867427010602619,
     "sustainable": true,
-    "buffer_seconds": 60.903726284450975,
+    "buffer_seconds": 60.098102654574994,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314763.96813613514,
-    "upload_bytes_per_second": 5209912.305353928,
+    "generation_bytes_per_second": 315188.36230490956,
+    "upload_bytes_per_second": 5378013.190800452,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.54102023184815,
+    "estimated_full_shards_per_hour": 67.63208534107652,
     "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": 2.659455514000001,
-      "bytes": 1230518,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1675947069998074,
       "bytes": 6176188,
@@ -562,6 +556,12 @@
       "bytes": 6045751,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.4661580640004104,
+      "bytes": 8583866,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -139,3 +139,4 @@
 {"step": 571000, "loss": 0.48727861046791077, "lr": 0.0003, "grad_norm": 0.18006728943270417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83577426920, "tokens_per_second": 148787.13178709097, "data_wait_s": 1.1567000910872594e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 571100, "loss": 0.5819798111915588, "lr": 0.0003, "grad_norm": 0.167887702044046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83592114214, "tokens_per_second": 153179.62478965265, "data_wait_s": 7.991000529727899e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 571200, "loss": 0.6157252192497253, "lr": 0.0003, "grad_norm": 0.16089026064670048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83606793094, "tokens_per_second": 153577.0429783949, "data_wait_s": 1.6521000361535698e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 571300, "loss": 0.6090176105499268, "lr": 0.0003, "grad_norm": 0.1407998120763945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83621563952, "tokens_per_second": 154195.777095314, "data_wait_s": 9.938999937730841e-06, "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:fd6aa22cc4c40b461c669b99a84485974f82b4d4fc6c8256e51204c5e41aa1d5
-size 301152
+oid sha256:f1295c3cd3ee1ec890c02fd2c1b202540209f9dc9906ed5dbcdcbc4e78849e0d
+size 303270