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": 77357.17567504723,
-      "required_pages_per_second_upper_estimate": 0.005351748906576695
+      "required_token_positions_per_second": 77232.21234640622,
+      "required_pages_per_second_upper_estimate": 0.005343103653546439
     },
     {
       "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": 9669.646959380903,
-      "required_pages_per_second_upper_estimate": 0.004048479256702594
+      "required_token_positions_per_second": 9654.026543300777,
+      "required_pages_per_second_upper_estimate": 0.004041939314681225
     },
     {
       "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": 19339.293918761807,
-      "required_pages_per_second_upper_estimate": 0.00402786491180066
+      "required_token_positions_per_second": 19308.053086601554,
+      "required_pages_per_second_upper_estimate": 0.004021358270337827
     },
     {
       "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": 29008.940878142712,
-      "required_pages_per_second_upper_estimate": 0.0017390417115566456
+      "required_token_positions_per_second": 28962.07962990233,
+      "required_pages_per_second_upper_estimate": 0.0017362324512776178
     },
     {
       "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": 29008.940878142712,
-      "required_pages_per_second_upper_estimate": 0.0022553175564295233
+      "required_token_positions_per_second": 28962.07962990233,
+      "required_pages_per_second_upper_estimate": 0.002251674300499681
     },
     {
       "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": 29008.940878142712,
-      "required_pages_per_second_upper_estimate": 0.3720955462108325
+      "required_token_positions_per_second": 28962.07962990233,
+      "required_pages_per_second_upper_estimate": 0.37149446043409307
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3772817625156506,
-    "producer_p95_seconds": 0.17666148299940687,
-    "gpu_update_seconds": 0.953085468240306,
-    "producer_batches_per_second": 2.650538932314592,
-    "gpu_batches_per_second": 1.0492238454189349,
-    "producer_headroom": 2.5261901393942137,
+    "producer_batch_seconds": 0.07369281603909883,
+    "producer_p95_seconds": 0.1989251239992882,
+    "gpu_update_seconds": 0.9546275803846078,
+    "producer_batches_per_second": 13.569843761560625,
+    "gpu_batches_per_second": 1.0475289217991295,
+    "producer_headroom": 12.954147116295784,
     "sustainable": true,
-    "buffer_seconds": 60.044384499139284,
+    "buffer_seconds": 60.14153756423029,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459719.4723678229,
+    "generation_bytes_per_second": 458427.8237504764,
     "upload_bytes_per_second": 5651389.22452136,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.64509704853072,
+    "estimated_full_shards_per_hour": 98.36793932328911,
     "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": [

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

@@ -95,3 +95,4 @@
 {"step": 566600, "loss": 0.5817695260047913, "lr": 0.0003, "grad_norm": 0.14382313820745526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82933121837, "tokens_per_second": 152959.16644425303, "data_wait_s": 1.3075999959255569e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 566700, "loss": 0.60850989818573, "lr": 0.0003, "grad_norm": 0.18116197041696708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82947853160, "tokens_per_second": 144644.81709420827, "data_wait_s": 1.2070000593666919e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 566800, "loss": 0.6198548674583435, "lr": 0.0003, "grad_norm": 0.1820872778762793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82962588143, "tokens_per_second": 153567.84994269232, "data_wait_s": 1.391200021316763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 566900, "loss": 0.5459545254707336, "lr": 0.0003, "grad_norm": 0.18065837711928423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82977431019, "tokens_per_second": 154521.72931054942, "data_wait_s": 1.286200131289661e-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:e8ee8cb685baec2651f9421b3ab58de2d9767a1432a8157353cf7db22dd6a0a8
-size 207177
+oid sha256:c33a5aa72a4141dd39f492d4ea6837002f2c2b11faca618f7d8d9133a06ee15e
+size 209295