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": 76476.41139722221,
-      "required_pages_per_second_upper_estimate": 0.005290815590182075
+      "required_token_positions_per_second": 75081.73992875044,
+      "required_pages_per_second_upper_estimate": 0.005194328981909546
     },
     {
       "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": 9559.551424652776,
-      "required_pages_per_second_upper_estimate": 0.004002384555368126
+      "required_token_positions_per_second": 9385.217491093805,
+      "required_pages_per_second_upper_estimate": 0.003929394577893494
     },
     {
       "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": 19119.102849305553,
-      "required_pages_per_second_upper_estimate": 0.003982004918861916
+      "required_token_positions_per_second": 18770.43498218761,
+      "required_pages_per_second_upper_estimate": 0.003909386597131241
     },
     {
       "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": 28678.654273958327,
-      "required_pages_per_second_upper_estimate": 0.0017192415339542353
+      "required_token_positions_per_second": 28155.652473281414,
+      "required_pages_per_second_upper_estimate": 0.001687888374581165
     },
     {
       "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": 28678.654273958327,
-      "required_pages_per_second_upper_estimate": 0.002229639225731425
+      "required_token_positions_per_second": 28155.652473281414,
+      "required_pages_per_second_upper_estimate": 0.002188978136170598
     },
     {
       "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": 28678.654273958327,
-      "required_pages_per_second_upper_estimate": 0.3678589842864808
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28155.652473281414,
+      "required_pages_per_second_upper_estimate": 0.36115047874297934
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 720750,
-    "producer_batch_seconds": 0.08880520189846663,
-    "producer_p95_seconds": 0.1874910499991529,
-    "gpu_update_seconds": 0.9640619722211229,
-    "producer_batches_per_second": 11.26060161592028,
-    "gpu_batches_per_second": 1.0372777153486086,
-    "producer_headroom": 10.855917802240468,
+    "producer_batch_seconds": 0.3875628712813892,
+    "producer_p95_seconds": 0.19505584299986367,
+    "gpu_update_seconds": 0.9819697847967418,
+    "producer_batches_per_second": 2.5802265234895323,
+    "gpu_batches_per_second": 1.0183612729051437,
+    "producer_headroom": 2.533704483997861,
     "sustainable": true,
-    "buffer_seconds": 60.73590424993074,
+    "buffer_seconds": 60.88212665739799,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337362.524992241,
-    "upload_bytes_per_second": 4081802.9223013143,
+    "generation_bytes_per_second": 337305.65279135725,
+    "upload_bytes_per_second": 4124460.623959151,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.39014446568892,
+    "estimated_full_shards_per_hour": 72.37794101529634,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 1163617,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 2.120087691999288,
+      "bytes": 11751236,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -93,3 +93,4 @@
 {"step": 809300, "loss": 0.537666916847229, "lr": 3e-05, "grad_norm": 0.26669529174821927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118457904094, "tokens_per_second": 149873.11519992395, "data_wait_s": 1.1402000382076949e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809400, "loss": 0.5624492168426514, "lr": 3e-05, "grad_norm": 0.27627468464037896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118472671545, "tokens_per_second": 152518.71777550018, "data_wait_s": 1.6554999092477374e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809500, "loss": 0.5444309711456299, "lr": 3e-05, "grad_norm": 0.2800186850872327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118487430031, "tokens_per_second": 151739.90095548856, "data_wait_s": 1.8564998754300177e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 809600, "loss": 0.4876198172569275, "lr": 3e-05, "grad_norm": 0.30805431226618685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118502218881, "tokens_per_second": 153219.5542393025, "data_wait_s": 1.749500006553717e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5f2a2bdc644de73deab0fe132ae889bbaf0d2afda8f7660ad32b94aaaa99e5d
-size 202158
+oid sha256:bb21f16d32e8af9d06b886d6ef3d10fa674b95acfcd5f8caf13c9edf54f834a5
+size 204276