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": 74977.59649181737,
-      "required_pages_per_second_upper_estimate": 0.006498868034929751
+      "required_token_positions_per_second": 74666.55993666522,
+      "required_pages_per_second_upper_estimate": 0.006471908174644104
     },
     {
       "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": 9372.199561477171,
-      "required_pages_per_second_upper_estimate": 0.0037387105319439807
+      "required_token_positions_per_second": 9333.319992083152,
+      "required_pages_per_second_upper_estimate": 0.003723200890411342
     },
     {
       "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": 18744.399122954343,
-      "required_pages_per_second_upper_estimate": 0.0043361868469380714
+      "required_token_positions_per_second": 18666.639984166304,
+      "required_pages_per_second_upper_estimate": 0.00431819863869356
     },
     {
       "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": 28116.598684431516,
-      "required_pages_per_second_upper_estimate": 0.0017121533434892495
+      "required_token_positions_per_second": 27999.959976249458,
+      "required_pages_per_second_upper_estimate": 0.0017050506581169696
     },
     {
       "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": 28116.598684431516,
-      "required_pages_per_second_upper_estimate": 0.0021644935984541486
+      "required_token_positions_per_second": 27999.959976249458,
+      "required_pages_per_second_upper_estimate": 0.0021555144278216846
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28116.598684431516,
-      "required_pages_per_second_upper_estimate": 0.34911405545813123
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27999.959976249458,
+      "required_pages_per_second_upper_estimate": 0.34766579306715495
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11077144393755134,
-    "producer_p95_seconds": 0.2758080399980827,
-    "gpu_update_seconds": 0.9833337350050458,
-    "producer_batches_per_second": 9.027597406455778,
-    "gpu_batches_per_second": 1.01694873713945,
-    "producer_headroom": 8.877141075812025,
+    "producer_batch_seconds": 0.07310005958578358,
+    "producer_p95_seconds": 0.2482572920016537,
+    "gpu_update_seconds": 0.9874299828804041,
+    "producer_batches_per_second": 13.67987941003647,
+    "gpu_batches_per_second": 1.0127300338631893,
+    "producer_headroom": 13.507923091658304,
     "sustainable": true,
-    "buffer_seconds": 60.96669157031284,
+    "buffer_seconds": 60.23322895570465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429583.7320080436,
-    "upload_bytes_per_second": 5784208.254538508,
+    "generation_bytes_per_second": 429473.8914452338,
+    "upload_bytes_per_second": 5721870.433251893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.17866869145375,
+    "estimated_full_shards_per_hour": 92.1550994636322,
     "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": 1.0268402850015264,
-      "bytes": 7548268,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.2450350950002758,
       "bytes": 5949961,
@@ -562,6 +556,12 @@
       "bytes": 6490741,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9587836459977552,
+      "bytes": 2593270,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 133900, "loss": 0.6582215577363968, "lr": 0.0003, "grad_norm": 0.14935984275103895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19595797030, "tokens_per_second": 150520.7337200414, "data_wait_s": 1.3793000107398257e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134000, "loss": 0.6526389606297016, "lr": 0.0003, "grad_norm": 0.1691735713082882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19610482155, "tokens_per_second": 150119.77806600573, "data_wait_s": 1.133299883804284e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134100, "loss": 0.5643828809261322, "lr": 0.0003, "grad_norm": 0.12965959476805836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19625249699, "tokens_per_second": 150435.3461799266, "data_wait_s": 9.978997695725411e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 134200, "loss": 0.6500321365892887, "lr": 0.0003, "grad_norm": 0.15373344914584586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19640017343, "tokens_per_second": 150744.41159973186, "data_wait_s": 1.6715002857381478e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d157361984d29bbf95905c81cf48ec744cd04143b1c658ecf4cc8740c78f2563
-size 407052
+oid sha256:798882fd2b31f5d5508633a0dd46d4bb5c006bcbe8ce08d06f034008a6087348
+size 409170