Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76377.8677482956,
-      "required_pages_per_second_upper_estimate": 0.0066202399984862294
+      "required_token_positions_per_second": 76433.27194739044,
+      "required_pages_per_second_upper_estimate": 0.006625042294043112
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9547.23346853695,
-      "required_pages_per_second_upper_estimate": 0.003808534174460248
+      "required_token_positions_per_second": 9554.158993423805,
+      "required_pages_per_second_upper_estimate": 0.003811296869883439
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19094.4669370739,
-      "required_pages_per_second_upper_estimate": 0.0044171688747515255
+      "required_token_positions_per_second": 19108.31798684761,
+      "required_pages_per_second_upper_estimate": 0.004420373071346523
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28641.70040561085,
-      "required_pages_per_second_upper_estimate": 0.0017441292833132573
+      "required_token_positions_per_second": 28662.476980271415,
+      "required_pages_per_second_upper_estimate": 0.0017453944677038229
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28641.70040561085,
-      "required_pages_per_second_upper_estimate": 0.0022049173825251307
+      "required_token_positions_per_second": 28662.476980271415,
+      "required_pages_per_second_upper_estimate": 0.002206516820755741
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28641.70040561085,
-      "required_pages_per_second_upper_estimate": 0.35563406143276816
+      "required_token_positions_per_second": 28662.476980271415,
+      "required_pages_per_second_upper_estimate": 0.35589203695533
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.39413603042157774,
-    "producer_p95_seconds": 0.3822081159960362,
-    "gpu_update_seconds": 0.9653058166401257,
-    "producer_batches_per_second": 2.5371950870119004,
-    "gpu_batches_per_second": 1.0359411315686795,
-    "producer_headroom": 2.4491691754433376,
+    "producer_batch_seconds": 0.08707029998436155,
+    "producer_p95_seconds": 0.3324667619963293,
+    "gpu_update_seconds": 0.964606095245373,
+    "producer_batches_per_second": 11.484972489811188,
+    "gpu_batches_per_second": 1.0366925991128262,
+    "producer_headroom": 11.0784744673973,
     "sustainable": true,
-    "buffer_seconds": 60.81426644832792,
+    "buffer_seconds": 60.7701840004585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402248.7431974242,
-    "upload_bytes_per_second": 5285435.157326066,
+    "generation_bytes_per_second": 402159.05934215855,
+    "upload_bytes_per_second": 5236678.518853753,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.31321641866727,
+    "estimated_full_shards_per_hour": 86.29397235106056,
     "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": 1.4456373529974371,
-      "bytes": 7578448,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.2644371399983356,
       "bytes": 12142497,
@@ -562,6 +556,12 @@
       "bytes": 12240192,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 0.9206129219965078,
+      "bytes": 852268,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -357,3 +357,4 @@
 {"step": 190700, "loss": 0.5992460511624813, "lr": 0.0003, "grad_norm": 0.14724762084252568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27910923120, "tokens_per_second": 151043.11295818255, "data_wait_s": 7.196998922154307e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 190800, "loss": 0.5666274484246969, "lr": 0.0003, "grad_norm": 0.13830076820219755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27925378730, "tokens_per_second": 148450.05855414632, "data_wait_s": 9.757997759152204e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 190900, "loss": 0.5981983579695225, "lr": 0.0003, "grad_norm": 0.149478788079187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27940072447, "tokens_per_second": 150759.55153794054, "data_wait_s": 9.71899862634018e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 191000, "loss": 0.6219041235744953, "lr": 0.0003, "grad_norm": 0.13180663164012466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27954752847, "tokens_per_second": 150940.91277868982, "data_wait_s": 8.448005246464163e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0de643aa42fc72752f4dc5623f3cd1f1f69521107bf5906bacc6fe5a7b78efcf
-size 766008
+oid sha256:7d7927b530e330348b8d18eaccb59bf6d218cccc634e09c2d0e4e2bf8507f85b
+size 768126