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": 77401.2943531824,
-      "required_pages_per_second_upper_estimate": 0.005354801139099523
+      "required_token_positions_per_second": 77339.50712285408,
+      "required_pages_per_second_upper_estimate": 0.005350526555139283
     },
     {
       "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": 9675.1617941478,
-      "required_pages_per_second_upper_estimate": 0.004050788202856648
+      "required_token_positions_per_second": 9667.43839035676,
+      "required_pages_per_second_upper_estimate": 0.00404755457497235
     },
     {
       "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": 19350.3235882956,
-      "required_pages_per_second_upper_estimate": 0.004030162101092603
+      "required_token_positions_per_second": 19334.87678071352,
+      "required_pages_per_second_upper_estimate": 0.004026944938432966
     },
     {
       "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": 29025.4853824434,
-      "required_pages_per_second_upper_estimate": 0.0017400335293274764
+      "required_token_positions_per_second": 29002.31517107028,
+      "required_pages_per_second_upper_estimate": 0.0017386445105345221
     },
     {
       "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": 29025.4853824434,
-      "required_pages_per_second_upper_estimate": 0.002256603818867318
+      "required_token_positions_per_second": 29002.31517107028,
+      "required_pages_per_second_upper_estimate": 0.002254802436848047
     },
     {
       "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": 29025.4853824434,
-      "required_pages_per_second_upper_estimate": 0.372307761347897
+      "required_token_positions_per_second": 29002.31517107028,
+      "required_pages_per_second_upper_estimate": 0.37201055875463734
     }
   ],
   "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": 0.3903470969531,
-    "producer_p95_seconds": 0.19451853999999003,
-    "gpu_update_seconds": 0.9525422102578655,
-    "producer_batches_per_second": 2.5618225620367543,
-    "gpu_batches_per_second": 1.0498222432886066,
-    "producer_headroom": 2.4402441255309575,
+    "producer_batch_seconds": 0.05293472189064374,
+    "producer_p95_seconds": 0.17497010199986107,
+    "gpu_update_seconds": 0.9533032048275509,
+    "producer_batches_per_second": 18.891192100071294,
+    "gpu_batches_per_second": 1.0489842003425305,
+    "producer_headroom": 18.009033972010876,
     "sustainable": true,
-    "buffer_seconds": 60.01015924624552,
+    "buffer_seconds": 60.05810190413571,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354599.8802363824,
-    "upload_bytes_per_second": 4281714.755142543,
+    "generation_bytes_per_second": 353529.44063540804,
+    "upload_bytes_per_second": 4245835.198740466,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.08887963598828,
+    "estimated_full_shards_per_hour": 75.85918821617776,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 11707710,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.9451075770002717,
+      "bytes": 1133265,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -63,3 +63,4 @@
 {"step": 446500, "loss": 0.537320613861084, "lr": 0.0003, "grad_norm": 0.16300642715305638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65353900404, "tokens_per_second": 152319.31360882826, "data_wait_s": 1.3916999705543276e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 446600, "loss": 0.5930716395378113, "lr": 0.0003, "grad_norm": 0.1700229495231917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65368591709, "tokens_per_second": 152180.17333898123, "data_wait_s": 1.6773000425018836e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 446700, "loss": 0.539176344871521, "lr": 0.0003, "grad_norm": 0.15264892739942929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65383176323, "tokens_per_second": 152023.39627403402, "data_wait_s": 1.4923999515303876e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 446800, "loss": 0.5781931281089783, "lr": 0.0003, "grad_norm": 0.14260654019291844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65397905096, "tokens_per_second": 152879.43145148747, "data_wait_s": 1.5163999705691822e-05, "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:f84c49358d1729f54ecec280004d7c55b774a56fdf8c636ade6f7c7529d6eadc
-size 138618
+oid sha256:13848589593edc82b293c906e49345b0ad82d30a2113a5bff240da3f381b1131
+size 140736