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": 76649.71717301355,
-      "required_pages_per_second_upper_estimate": 0.005302805285876121
+      "required_token_positions_per_second": 77335.73501713686,
+      "required_pages_per_second_upper_estimate": 0.005350265592113275
     },
     {
       "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": 9581.214646626693,
-      "required_pages_per_second_upper_estimate": 0.004011454494029089
+      "required_token_positions_per_second": 9666.966877142107,
+      "required_pages_per_second_upper_estimate": 0.004047357162235691
     },
     {
       "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": 19162.429293253386,
-      "required_pages_per_second_upper_estimate": 0.003991028674541087
+      "required_token_positions_per_second": 19333.933754284215,
+      "required_pages_per_second_upper_estimate": 0.004026748530897021
     },
     {
       "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": 28743.64393988008,
-      "required_pages_per_second_upper_estimate": 0.0017231375651927702
+      "required_token_positions_per_second": 29000.900631426324,
+      "required_pages_per_second_upper_estimate": 0.001738559711042744
     },
     {
       "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": 28743.64393988008,
-      "required_pages_per_second_upper_estimate": 0.0022346918863975114
+      "required_token_positions_per_second": 29000.900631426324,
+      "required_pages_per_second_upper_estimate": 0.0022546924626126318
     },
     {
       "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": 28743.64393988008,
-      "required_pages_per_second_upper_estimate": 0.36869260194045844
+      "required_token_positions_per_second": 29000.900631426324,
+      "required_pages_per_second_upper_estimate": 0.371992414558899
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11052775129718384,
-    "producer_p95_seconds": 0.33612018099665875,
-    "gpu_update_seconds": 0.9618822184768321,
-    "producer_batches_per_second": 9.047501539330414,
-    "gpu_batches_per_second": 1.03962832537182,
-    "producer_headroom": 8.702630852323692,
+    "producer_batch_seconds": 0.34044047444552916,
+    "producer_p95_seconds": 0.34030860100028804,
+    "gpu_update_seconds": 0.9533497028723212,
+    "producer_batches_per_second": 2.9373710679633103,
+    "gpu_batches_per_second": 1.0489330378843433,
+    "producer_headroom": 2.8003418348685747,
     "sustainable": true,
-    "buffer_seconds": 60.59857976404042,
+    "buffer_seconds": 60.061031280956236,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 487685.6556864421,
-    "upload_bytes_per_second": 5056775.41149406,
+    "generation_bytes_per_second": 467961.2677016521,
+    "upload_bytes_per_second": 4965568.333143834,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.64598896927784,
+    "estimated_full_shards_per_hour": 100.41359446799441,
     "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": 1.119130936000147,
-      "bytes": 7848522,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.0408579100003408,
       "bytes": 6882526,
@@ -562,6 +556,12 @@
       "bytes": 6203770,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.3150392690004082,
+      "bytes": 914685,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -383,3 +383,4 @@
 {"step": 683300, "loss": 0.5685223937034607, "lr": 0.0001, "grad_norm": 0.20739122870950957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100017359825, "tokens_per_second": 152272.69776061346, "data_wait_s": 1.486000110162422e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683400, "loss": 0.5565347075462341, "lr": 0.0001, "grad_norm": 0.1834445118323151, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100032050677, "tokens_per_second": 152632.3169792421, "data_wait_s": 1.0711002687457949e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683500, "loss": 0.49607446789741516, "lr": 0.0001, "grad_norm": 0.19096895211888984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100046773572, "tokens_per_second": 154159.41512214197, "data_wait_s": 1.0621995897963643e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 683600, "loss": 0.5936766862869263, "lr": 0.0001, "grad_norm": 0.21774702035682403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100061483139, "tokens_per_second": 153687.8288892015, "data_wait_s": 1.3689998013433069e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7ac798f78e51ff52749022b20c4a0b3804c649c25bea57cce2f5a6f4801177d
-size 821076
+oid sha256:28fb0df40fb543ad50c62ffd0dcd1c7aec8b182ba48f2a3f363e8a5cc75bbba1
+size 823194