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": 77234.02274089694,
-      "required_pages_per_second_upper_estimate": 0.005343228900838001
+      "required_token_positions_per_second": 76577.03203210907,
+      "required_pages_per_second_upper_estimate": 0.005297776759175587
     },
     {
       "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": 9654.252842612117,
-      "required_pages_per_second_upper_estimate": 0.004042034061477216
+      "required_token_positions_per_second": 9572.129004013634,
+      "required_pages_per_second_upper_estimate": 0.004007650525196793
     },
     {
       "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": 19308.505685224234,
-      "required_pages_per_second_upper_estimate": 0.0040214525346951024
+      "required_token_positions_per_second": 19144.258008027267,
+      "required_pages_per_second_upper_estimate": 0.003987244075037503
     },
     {
       "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": 28962.758527836348,
-      "required_pages_per_second_upper_estimate": 0.0017362731501721449
+      "required_token_positions_per_second": 28716.3870120409,
+      "required_pages_per_second_upper_estimate": 0.0017215035539877306
     },
     {
       "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": 28962.758527836348,
-      "required_pages_per_second_upper_estimate": 0.0022517270818278945
+      "required_token_positions_per_second": 28716.3870120409,
+      "required_pages_per_second_upper_estimate": 0.0022325727801485703
     },
     {
       "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": 28962.758527836348,
-      "required_pages_per_second_upper_estimate": 0.3715031686078468
+      "required_token_positions_per_second": 28716.3870120409,
+      "required_pages_per_second_upper_estimate": 0.3683429793363464
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08834054339848763,
-    "producer_p95_seconds": 0.3257269039968378,
-    "gpu_update_seconds": 0.9546052035557068,
-    "producer_batches_per_second": 11.31983075414408,
-    "gpu_batches_per_second": 1.0475534768459327,
-    "producer_headroom": 10.805969341275858,
+    "producer_batch_seconds": 0.6381294971565126,
+    "producer_p95_seconds": 0.33471313200425357,
+    "gpu_update_seconds": 0.962795214746447,
+    "producer_batches_per_second": 1.5670800432450975,
+    "gpu_batches_per_second": 1.0386424700535626,
+    "producer_headroom": 1.508777166761035,
     "sustainable": true,
-    "buffer_seconds": 60.140127824009525,
+    "buffer_seconds": 60.65609852902616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424718.82513725443,
-    "upload_bytes_per_second": 5318501.385929223,
+    "generation_bytes_per_second": 359985.80400561955,
+    "upload_bytes_per_second": 5379359.971880107,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.13477292621828,
+    "estimated_full_shards_per_hour": 77.24457349897804,
     "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": 1.3229683749996184,
-      "bytes": 6064094,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 0.8354571769996255,
       "bytes": 2427677,
@@ -562,6 +556,12 @@
       "bytes": 1129072,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.0205191320055746,
+      "bytes": 9233133,
+      "pages": 13,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -339,3 +339,4 @@
 {"step": 474100, "loss": 0.6103756427764893, "lr": 0.0003, "grad_norm": 0.1460226722104052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69393242883, "tokens_per_second": 149687.91178289353, "data_wait_s": 1.7116006347350776e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 474200, "loss": 0.5817875862121582, "lr": 0.0003, "grad_norm": 0.15582655146374855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69407863053, "tokens_per_second": 152170.59249969205, "data_wait_s": 1.9254999642726034e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 474300, "loss": 0.5846430659294128, "lr": 0.0003, "grad_norm": 0.15215446976376132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69422543844, "tokens_per_second": 152182.48746607485, "data_wait_s": 1.353699917672202e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 474400, "loss": 0.6118581891059875, "lr": 0.0003, "grad_norm": 0.16554137706445823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69437260186, "tokens_per_second": 113874.5264129922, "data_wait_s": 9.49700188357383e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12beed73c9a338f575c998db272376696f2d85bd7c2fa2da5a9c5cb818b93951
-size 727101
+oid sha256:52cf75a9ad3387dd450d0944733a8a6db619cea439ab5976a0c534df9a8a4885
+size 729219