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": 76508.19557523186,
-      "required_pages_per_second_upper_estimate": 0.005293014493366236
+      "required_token_positions_per_second": 76515.26714994019,
+      "required_pages_per_second_upper_estimate": 0.005293503721312872
     },
     {
       "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": 9563.524446903983,
-      "required_pages_per_second_upper_estimate": 0.004004047976818567
+      "required_token_positions_per_second": 9564.408393742524,
+      "required_pages_per_second_upper_estimate": 0.004004418066900955
     },
     {
       "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": 19127.048893807965,
-      "required_pages_per_second_upper_estimate": 0.0039836598703804835
+      "required_token_positions_per_second": 19128.816787485048,
+      "required_pages_per_second_upper_estimate": 0.003984028076010928
     },
     {
       "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": 28690.573340711944,
-      "required_pages_per_second_upper_estimate": 0.0017199560638067534
+      "required_token_positions_per_second": 28693.22518122757,
+      "required_pages_per_second_upper_estimate": 0.0017201150375965393
     },
     {
       "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": 28690.573340711944,
-      "required_pages_per_second_upper_estimate": 0.002230565880745085
+      "required_token_positions_per_second": 28693.22518122757,
+      "required_pages_per_second_upper_estimate": 0.002230772049680961
     },
     {
       "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": 28690.573340711944,
-      "required_pages_per_second_upper_estimate": 0.3680118692770994
+      "required_token_positions_per_second": 28693.22518122757,
+      "required_pages_per_second_upper_estimate": 0.36804588423990936
     }
   ],
   "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": 0.077090419390629,
-    "producer_p95_seconds": 0.22068424499957473,
-    "gpu_update_seconds": 0.9636614671888577,
-    "producer_batches_per_second": 12.971780513125067,
-    "gpu_batches_per_second": 1.0377088158532966,
-    "producer_headroom": 12.500405041329936,
+    "producer_batch_seconds": 0.2878900047579265,
+    "producer_p95_seconds": 0.22802923299968825,
+    "gpu_update_seconds": 0.963572405171399,
+    "producer_batches_per_second": 3.4735488675296464,
+    "gpu_batches_per_second": 1.037804730223798,
+    "producer_headroom": 3.3470158367659306,
     "sustainable": true,
-    "buffer_seconds": 60.710672432898036,
+    "buffer_seconds": 60.705061525798136,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332059.11640745215,
-    "upload_bytes_per_second": 5328368.002377797,
+    "generation_bytes_per_second": 309354.06131486606,
+    "upload_bytes_per_second": 5376599.606311192,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.25215644042657,
+    "estimated_full_shards_per_hour": 66.38018016418921,
     "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": 0.9982179830003588,
-      "bytes": 6432855,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 2.2747711609999897,
       "bytes": 6302843,
@@ -562,6 +556,12 @@
       "bytes": 2235467,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.0548529980005696,
+      "bytes": 10543433,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 409600, "loss": 0.5977052450180054, "lr": 0.0003, "grad_norm": 0.17293401276413004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59954720288, "tokens_per_second": 152388.0654397513, "data_wait_s": 9.227000191458501e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 409700, "loss": 0.6045535206794739, "lr": 0.0003, "grad_norm": 0.1535792509856481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59969279489, "tokens_per_second": 148243.25800603238, "data_wait_s": 1.4062999980524182e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 409800, "loss": 0.54542475938797, "lr": 0.0003, "grad_norm": 0.1634236049966163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59983983537, "tokens_per_second": 152699.83598598043, "data_wait_s": 1.631800114409998e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 409900, "loss": 0.5718177556991577, "lr": 0.0003, "grad_norm": 0.139769420556045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59998617478, "tokens_per_second": 152259.28878810006, "data_wait_s": 9.601999408914708e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7b726dc4a144c9577ae9f996f501630e337fd4f88dcbaa8c7ad48df99f3f670
-size 270717
+oid sha256:1fae2a17f8e6c4aa47e6ad0efc056440bb3260f0a6052211dd5a63bd991b0407
+size 272835