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": 76447.04928907607,
-      "required_pages_per_second_upper_estimate": 0.005288784251411047
+      "required_token_positions_per_second": 76133.86790569121,
+      "required_pages_per_second_upper_estimate": 0.005267117636627556
     },
     {
       "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": 9555.881161134508,
-      "required_pages_per_second_upper_estimate": 0.004000847892676845
+      "required_token_positions_per_second": 9516.733488211401,
+      "required_pages_per_second_upper_estimate": 0.003984457579520311
     },
     {
       "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": 19111.762322269016,
-      "required_pages_per_second_upper_estimate": 0.003980476080662922
+      "required_token_positions_per_second": 19033.466976422802,
+      "required_pages_per_second_upper_estimate": 0.003964169224910276
     },
     {
       "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": 28667.64348340352,
-      "required_pages_per_second_upper_estimate": 0.0017185814538729783
+      "required_token_positions_per_second": 28550.200464634203,
+      "required_pages_per_second_upper_estimate": 0.0017115409242228641
     },
     {
       "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": 28667.64348340352,
-      "required_pages_per_second_upper_estimate": 0.002228783185197138
+      "required_token_positions_per_second": 28550.200464634203,
+      "required_pages_per_second_upper_estimate": 0.0022196525070650695
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28667.64348340352,
-      "required_pages_per_second_upper_estimate": 0.3677177496877095
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28550.200464634203,
+      "required_pages_per_second_upper_estimate": 0.3662113167434256
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1437659510860101,
-    "producer_p95_seconds": 0.2566920350000146,
-    "gpu_update_seconds": 0.964432253247679,
-    "producer_batches_per_second": 6.955749900765692,
-    "gpu_batches_per_second": 1.0368794662689353,
-    "producer_headroom": 6.708349549822776,
+    "producer_batch_seconds": 0.08448628837501815,
+    "producer_p95_seconds": 0.24035928600278567,
+    "gpu_update_seconds": 0.9683995050839738,
+    "producer_batches_per_second": 11.836240166702497,
+    "gpu_batches_per_second": 1.0326316718979385,
+    "producer_headroom": 11.46220911948975,
     "sustainable": true,
-    "buffer_seconds": 60.75923195460378,
+    "buffer_seconds": 60.040769315206376,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322944.1373505963,
-    "upload_bytes_per_second": 4903172.445268796,
+    "generation_bytes_per_second": 337316.8108625343,
+    "upload_bytes_per_second": 4892439.762586308,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.29629412067811,
+    "estimated_full_shards_per_hour": 72.38033527762433,
     "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": 1.0955314789989643,
-      "bytes": 11552086,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.7776427819990204,
       "bytes": 236866,
@@ -562,6 +556,12 @@
       "bytes": 2322033,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2023178439994808,
+      "bytes": 11142768,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 462700, "loss": 0.5562172532081604, "lr": 0.0003, "grad_norm": 0.15023338817400747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67725234326, "tokens_per_second": 151000.43977930836, "data_wait_s": 1.2661999789997935e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 462800, "loss": 0.5868706107139587, "lr": 0.0003, "grad_norm": 0.16302416748524298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67739832563, "tokens_per_second": 151067.1121372493, "data_wait_s": 7.637001544935629e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 462900, "loss": 0.5904766917228699, "lr": 0.0003, "grad_norm": 0.18589227614191275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67754596057, "tokens_per_second": 154393.3384983321, "data_wait_s": 1.169900133390911e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 463000, "loss": 0.5758638381958008, "lr": 0.0003, "grad_norm": 0.16824982275587075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67769284022, "tokens_per_second": 152527.59354923086, "data_wait_s": 1.737300044624135e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95dc2a1ef105c2da4d934f5401d930553b7906adf732d56448cbd91b80c6cdf1
-size 484083
+oid sha256:e5e8666eae83b7e8b5209ae9c5e12781a7c0b85e2e5f80bf069bfdf6da377b61
+size 486201