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": 77188.74554330959,
-      "required_pages_per_second_upper_estimate": 0.005340096519251336
+      "required_token_positions_per_second": 77250.23767636965,
+      "required_pages_per_second_upper_estimate": 0.005344350687697837
     },
     {
       "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": 9648.593192913699,
-      "required_pages_per_second_upper_estimate": 0.004039664484335413
+      "required_token_positions_per_second": 9656.279709546207,
+      "required_pages_per_second_upper_estimate": 0.004042882668336725
     },
     {
       "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": 19297.186385827397,
-      "required_pages_per_second_upper_estimate": 0.0040190950231407405
+      "required_token_positions_per_second": 19312.559419092413,
+      "required_pages_per_second_upper_estimate": 0.0040222968205556936
     },
     {
       "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": 28945.779578741098,
-      "required_pages_per_second_upper_estimate": 0.001735255288099239
+      "required_token_positions_per_second": 28968.83912863862,
+      "required_pages_per_second_upper_estimate": 0.0017366376729057013
     },
     {
       "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": 28945.779578741098,
-      "required_pages_per_second_upper_estimate": 0.002250407043218243
+      "required_token_positions_per_second": 28968.83912863862,
+      "required_pages_per_second_upper_estimate": 0.002252199821794535
     },
     {
       "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": 28945.779578741098,
-      "required_pages_per_second_upper_estimate": 0.37128538087942814
+      "required_token_positions_per_second": 28968.83912863862,
+      "required_pages_per_second_upper_estimate": 0.3715811640260979
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.055880572890700364,
-    "producer_p95_seconds": 0.18779470600020431,
-    "gpu_update_seconds": 0.9551651536898237,
-    "producer_batches_per_second": 17.895306870886067,
-    "gpu_batches_per_second": 1.0469393655505317,
-    "producer_headroom": 17.092973537656448,
+    "producer_batch_seconds": 0.0739088230078977,
+    "producer_p95_seconds": 0.17580693800118752,
+    "gpu_update_seconds": 0.9544048305569539,
+    "producer_batches_per_second": 13.530184344745184,
+    "gpu_batches_per_second": 1.0477734059837465,
+    "producer_headroom": 12.913273296950877,
     "sustainable": true,
-    "buffer_seconds": 60.17540468245889,
+    "buffer_seconds": 60.1275043250881,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363317.8394954791,
-    "upload_bytes_per_second": 3075968.51152946,
+    "generation_bytes_per_second": 361195.247056976,
+    "upload_bytes_per_second": 3103863.2279518405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.9595507492855,
+    "estimated_full_shards_per_hour": 77.50409182340584,
     "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": [
@@ -430,6 +430,12 @@
       "bytes": 7623870,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1467835090006702,
+      "bytes": 6065504,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -84,3 +84,4 @@
 {"step": 653400, "loss": 0.5343234539031982, "lr": 0.0001, "grad_norm": 0.17442434352286854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95640817371, "tokens_per_second": 153838.3333628902, "data_wait_s": 1.3274000593810342e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 653500, "loss": 0.5341771841049194, "lr": 0.0001, "grad_norm": 0.2140702360987653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95655384593, "tokens_per_second": 151370.2923929644, "data_wait_s": 1.1362000805092975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 653600, "loss": 0.5025205016136169, "lr": 0.0001, "grad_norm": 0.1855704622409398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95669986827, "tokens_per_second": 152760.85696883049, "data_wait_s": 1.1673999324557371e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 653700, "loss": 0.5879846215248108, "lr": 0.0001, "grad_norm": 0.19696340840845797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95684610244, "tokens_per_second": 153054.70798256705, "data_wait_s": 1.1722000635927543e-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:f6e56c4617b6c7b6f501f3f639eb35a1a8e85836c2da693994fa05221b199091
-size 183096
+oid sha256:ef7ccfd328e8b21b77240230f05a7955bd54e1af3a9bcf22d9b213990154ee28
+size 185214