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": 77358.53680206041,
-      "required_pages_per_second_upper_estimate": 0.005351843072501717
+      "required_token_positions_per_second": 77322.91821283818,
+      "required_pages_per_second_upper_estimate": 0.005349378895852809
     },
     {
       "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": 9669.817100257551,
-      "required_pages_per_second_upper_estimate": 0.0040485504911347
+      "required_token_positions_per_second": 9665.364776604772,
+      "required_pages_per_second_upper_estimate": 0.0040466863961963724
     },
     {
       "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": 19339.634200515102,
-      "required_pages_per_second_upper_estimate": 0.004027935783516038
+      "required_token_positions_per_second": 19330.729553209545,
+      "required_pages_per_second_upper_estimate": 0.004026081180313632
     },
     {
       "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": 29009.45130077265,
-      "required_pages_per_second_upper_estimate": 0.00173907231061394
+      "required_token_positions_per_second": 28996.094329814314,
+      "required_pages_per_second_upper_estimate": 0.0017382715805999017
     },
     {
       "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": 29009.45130077265,
-      "required_pages_per_second_upper_estimate": 0.002255357239543889
+      "required_token_positions_per_second": 28996.094329814314,
+      "required_pages_per_second_upper_estimate": 0.0022543187938029847
     },
     {
       "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": 29009.45130077265,
-      "required_pages_per_second_upper_estimate": 0.3721020933642802
+      "required_token_positions_per_second": 28996.094329814314,
+      "required_pages_per_second_upper_estimate": 0.37193076448242474
     }
   ],
   "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": 0.12207786042961288,
-    "producer_p95_seconds": 0.23702635500012548,
-    "gpu_update_seconds": 0.9530686986576548,
-    "producer_batches_per_second": 8.191493498336461,
-    "gpu_batches_per_second": 1.049242306885585,
-    "producer_headroom": 7.807056048522171,
+    "producer_batch_seconds": 0.08070119609368476,
+    "producer_p95_seconds": 0.2094935360000818,
+    "gpu_update_seconds": 0.9535077271276435,
+    "producer_batches_per_second": 12.391390071085386,
+    "gpu_batches_per_second": 1.0487591988503442,
+    "producer_headroom": 11.815286182632676,
     "sustainable": true,
-    "buffer_seconds": 60.04332801543225,
+    "buffer_seconds": 60.07098680904154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325017.4339139021,
-    "upload_bytes_per_second": 5902139.219232914,
+    "generation_bytes_per_second": 354589.79293780285,
+    "upload_bytes_per_second": 5940450.847592103,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.74117529928968,
+    "estimated_full_shards_per_hour": 76.0867151365334,
     "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": 1.115209474999574,
-      "bytes": 812365,
-      "pages": 23,
-      "shards": 1
-    },
     {
       "seconds": 1.1693048819997784,
       "bytes": 7948420,
@@ -562,6 +556,12 @@
       "bytes": 5667842,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9565514620007889,
+      "bytes": 2665835,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 499400, "loss": 0.5696477890014648, "lr": 0.0003, "grad_norm": 0.1744138905646286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73095317497, "tokens_per_second": 151790.53945064975, "data_wait_s": 9.437000699108467e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 499500, "loss": 0.5315882563591003, "lr": 0.0003, "grad_norm": 0.13697746125448093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73109944416, "tokens_per_second": 152287.63724088288, "data_wait_s": 1.1536001693457365e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 499600, "loss": 0.5783843994140625, "lr": 0.0003, "grad_norm": 0.14725862291518543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73124620601, "tokens_per_second": 152956.6464081035, "data_wait_s": 1.386900112265721e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 499700, "loss": 0.6527587175369263, "lr": 0.0003, "grad_norm": 0.22528530225703736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73139199520, "tokens_per_second": 151710.46077142513, "data_wait_s": 1.0285999451298267e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d5c43542292b8b95a43a94fd25ed87716b551655166885baba612722309e9c9
-size 310959
+oid sha256:3c7ebb0d90b8a50337fcbb5a017df93b5016e8ff774c74c98716b603c1f5acfa
+size 313077