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": 74493.72727926374,
-      "required_pages_per_second_upper_estimate": 0.006456927477940909
+      "required_token_positions_per_second": 75387.9048395614,
+      "required_pages_per_second_upper_estimate": 0.006534432522595202
     },
     {
       "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": 9311.715909907967,
-      "required_pages_per_second_upper_estimate": 0.0037145826990218475
+      "required_token_positions_per_second": 9423.488104945174,
+      "required_pages_per_second_upper_estimate": 0.003759170298765428
     },
     {
       "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": 18623.431819815934,
-      "required_pages_per_second_upper_estimate": 0.0043082031903088226
+      "required_token_positions_per_second": 18846.97620989035,
+      "required_pages_per_second_upper_estimate": 0.004359916250705645
     },
     {
       "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": 27935.1477297239,
-      "required_pages_per_second_upper_estimate": 0.0017011039323471407
+      "required_token_positions_per_second": 28270.46431483552,
+      "required_pages_per_second_upper_estimate": 0.00172152295311028
     },
     {
       "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": 27935.1477297239,
-      "required_pages_per_second_upper_estimate": 0.002150525001672366
+      "required_token_positions_per_second": 28270.46431483552,
+      "required_pages_per_second_upper_estimate": 0.0021763386006099744
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27935.1477297239,
-      "required_pages_per_second_upper_estimate": 0.34686104187794303
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28270.46431483552,
+      "required_pages_per_second_upper_estimate": 0.3510245516326101
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12748910839042082,
-    "producer_p95_seconds": 0.36683067199919606,
-    "gpu_update_seconds": 0.9897209160122549,
-    "producer_batches_per_second": 7.843807307347498,
-    "gpu_batches_per_second": 1.010385840918833,
-    "producer_headroom": 7.763180153251583,
+    "producer_batch_seconds": 0.1433877157965071,
+    "producer_p95_seconds": 0.365623609002796,
+    "gpu_update_seconds": 0.9779818149463901,
+    "producer_batches_per_second": 6.974098125805836,
+    "gpu_batches_per_second": 1.0225139002761692,
+    "producer_headroom": 6.820541142689809,
     "sustainable": true,
-    "buffer_seconds": 60.372975876747546,
+    "buffer_seconds": 60.63487252667619,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 275307.78700229723,
-    "upload_bytes_per_second": 5197563.676997983,
+    "generation_bytes_per_second": 286633.755550721,
+    "upload_bytes_per_second": 5217232.626746375,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.07464225341499,
+    "estimated_full_shards_per_hour": 61.50493144885276,
     "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": 1.3748210780031513,
-      "bytes": 6494530,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9434360529994592,
       "bytes": 2587569,
@@ -562,6 +556,12 @@
       "bytes": 16141114,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.301262590000988,
+      "bytes": 7677650,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -346,3 +346,4 @@
 {"step": 271700, "loss": 0.6076798718422651, "lr": 0.0003, "grad_norm": 0.15424145714169504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39768594226, "tokens_per_second": 148545.56511140778, "data_wait_s": 1.3092001609038562e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 271800, "loss": 0.6356411464512348, "lr": 0.0003, "grad_norm": 0.15618310103831265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39783118487, "tokens_per_second": 138314.5787115614, "data_wait_s": 1.1853997420985252e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 271900, "loss": 0.5737604964524508, "lr": 0.0003, "grad_norm": 0.1566306782403431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39797576854, "tokens_per_second": 146978.6077398846, "data_wait_s": 1.3413999113254249e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 272000, "loss": 0.5325260702520609, "lr": 0.0003, "grad_norm": 0.15508612714968298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39812260420, "tokens_per_second": 148624.71231940892, "data_wait_s": 1.0889998520724475e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:efc99caa573ccbe1b42fa91beabbfab9042839ec4d3250a9b7c4436d58b9fca0
-size 742710
+oid sha256:49f3c346c380cea906def809b0283955719b5dcfc22edd9febca2604a2e03d7d
+size 744828