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": 76056.2065065976,
-      "required_pages_per_second_upper_estimate": 0.006592359217298812
+      "required_token_positions_per_second": 76211.15038048654,
+      "required_pages_per_second_upper_estimate": 0.006605789359585846
     },
     {
       "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": 9507.0258133247,
-      "required_pages_per_second_upper_estimate": 0.0037924947396380648
+      "required_token_positions_per_second": 9526.393797560817,
+      "required_pages_per_second_upper_estimate": 0.0038002209181270214
     },
     {
       "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": 19014.0516266494,
-      "required_pages_per_second_upper_estimate": 0.004398566207945945
+      "required_token_positions_per_second": 19052.787595121634,
+      "required_pages_per_second_upper_estimate": 0.004407527092522234
     },
     {
       "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": 28521.077439974102,
-      "required_pages_per_second_upper_estimate": 0.00173678397756577
+      "required_token_positions_per_second": 28579.18139268245,
+      "required_pages_per_second_upper_estimate": 0.0017403222034378285
     },
     {
       "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": 28521.077439974102,
-      "required_pages_per_second_upper_estimate": 0.0021956314927246744
+      "required_token_positions_per_second": 28579.18139268245,
+      "required_pages_per_second_upper_estimate": 0.0022001044958461983
     },
     {
       "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": 28521.077439974102,
-      "required_pages_per_second_upper_estimate": 0.35413632789865657
+      "required_token_positions_per_second": 28579.18139268245,
+      "required_pages_per_second_upper_estimate": 0.3548577845298739
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08393217017172105,
-    "producer_p95_seconds": 0.25809267300064676,
-    "gpu_update_seconds": 0.969388342996102,
-    "producer_batches_per_second": 11.914382744471514,
-    "gpu_batches_per_second": 1.0315783217583225,
-    "producer_headroom": 11.54966374648459,
+    "producer_batch_seconds": 0.09340655845302592,
+    "producer_p95_seconds": 0.20824329999959446,
+    "gpu_update_seconds": 0.9674174924786028,
+    "producer_batches_per_second": 10.705886359177864,
+    "gpu_batches_per_second": 1.0336798825478317,
+    "producer_headroom": 10.357061736356727,
     "sustainable": true,
-    "buffer_seconds": 60.102077265758325,
+    "buffer_seconds": 60.947302026151974,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 561254.2974125833,
-    "upload_bytes_per_second": 6198286.288939025,
+    "generation_bytes_per_second": 586667.641708753,
+    "upload_bytes_per_second": 6073830.604306104,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.21605344668924,
+    "estimated_full_shards_per_hour": 62.94260949347945,
     "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": 1.1538401379993957,
-      "bytes": 11570054,
-      "pages": 22,
-      "shards": 3
-    },
     {
       "seconds": 0.7532719520004321,
       "bytes": 2586371,
@@ -562,6 +556,12 @@
       "bytes": 6056576,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.102731920998849,
+      "bytes": 2598092,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -127,3 +127,4 @@
 {"step": 52700, "loss": 0.6120903808623552, "lr": 0.0003, "grad_norm": 0.14055092120346682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7713113620, "tokens_per_second": 150284.65807365673, "data_wait_s": 1.345999953628052e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 52800, "loss": 0.6057971864938736, "lr": 0.0003, "grad_norm": 0.14566707485955477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7727824093, "tokens_per_second": 150764.41672000327, "data_wait_s": 1.2390000847517513e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 52900, "loss": 0.6748346481472254, "lr": 0.0003, "grad_norm": 0.12579535463662597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7742525265, "tokens_per_second": 150655.0856055337, "data_wait_s": 1.0423998901387677e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 53000, "loss": 0.7239194959402084, "lr": 0.0003, "grad_norm": 0.1563958009252826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7757197901, "tokens_per_second": 150552.44047842812, "data_wait_s": 1.190599869005382e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe693a86d07245c51393becfcd5e076a4e2d78db8007d787ad008b3313737089
-size 274953
+oid sha256:9fc8f28ee57763d8ebf3af4f0fb09a4206c43fd8445e2716ee1261dc778705c2
+size 277071