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": 75882.02899442811,
-      "required_pages_per_second_upper_estimate": 0.005249694836399594
+      "required_token_positions_per_second": 76422.60943566434,
+      "required_pages_per_second_upper_estimate": 0.005287093445643752
     },
     {
       "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": 9485.253624303514,
-      "required_pages_per_second_upper_estimate": 0.00397127761787639
+      "required_token_positions_per_second": 9552.826179458043,
+      "required_pages_per_second_upper_estimate": 0.003999568835644181
     },
     {
       "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": 18970.507248607028,
-      "required_pages_per_second_upper_estimate": 0.003951056373965927
+      "required_token_positions_per_second": 19105.652358916086,
+      "required_pages_per_second_upper_estimate": 0.003979203536427075
     },
     {
       "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": 28455.76087291054,
-      "required_pages_per_second_upper_estimate": 0.0017058794148999377
+      "required_token_positions_per_second": 28658.47853837413,
+      "required_pages_per_second_upper_estimate": 0.001718032029412527
     },
     {
       "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": 28455.76087291054,
-      "required_pages_per_second_upper_estimate": 0.0022123102442044183
+      "required_token_positions_per_second": 28658.47853837413,
+      "required_pages_per_second_upper_estimate": 0.0022280706510334355
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28455.76087291054,
-      "required_pages_per_second_upper_estimate": 0.36499994706212774
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28658.47853837413,
+      "required_pages_per_second_upper_estimate": 0.3676001916134238
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0827029607655021,
-    "producer_p95_seconds": 0.21971883299920592,
-    "gpu_update_seconds": 0.9716134502072121,
-    "producer_batches_per_second": 12.091465538161605,
-    "gpu_batches_per_second": 1.0292158880537667,
-    "producer_headroom": 11.748230549594803,
+    "producer_batch_seconds": 0.2541760841795906,
+    "producer_p95_seconds": 0.2388652240006195,
+    "gpu_update_seconds": 0.9647406774570715,
+    "producer_batches_per_second": 3.9342804545428445,
+    "gpu_batches_per_second": 1.0365479795418884,
+    "producer_headroom": 3.795560391021779,
     "sustainable": true,
-    "buffer_seconds": 60.24003391284715,
+    "buffer_seconds": 60.7786626797955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472983.33726951666,
-    "upload_bytes_per_second": 3933022.686359951,
+    "generation_bytes_per_second": 448277.4506157306,
+    "upload_bytes_per_second": 3886909.442531961,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.49121368946194,
+    "estimated_full_shards_per_hour": 96.18990553716601,
     "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": 1.9842258310000034,
-      "bytes": 13303551,
-      "pages": 43,
-      "shards": 4
-    },
     {
       "seconds": 1.4946416839993617,
       "bytes": 5892242,
@@ -562,6 +556,12 @@
       "bytes": 1048131,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3689376410002296,
+      "bytes": 5834070,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 814700, "loss": 0.5845077633857727, "lr": 3e-05, "grad_norm": 0.28333416091779967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119249129344, "tokens_per_second": 152526.47696780687, "data_wait_s": 1.3072000001557171e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 814800, "loss": 0.5097643733024597, "lr": 3e-05, "grad_norm": 0.2702926412809111, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119263684066, "tokens_per_second": 149738.98734738777, "data_wait_s": 7.728000127826817e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 814900, "loss": 0.5868908166885376, "lr": 3e-05, "grad_norm": 0.2968352698170268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119278161782, "tokens_per_second": 149856.9553760005, "data_wait_s": 2.10089983738726e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 815000, "loss": 0.513067364692688, "lr": 3e-05, "grad_norm": 0.2654177458091664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119292759153, "tokens_per_second": 151537.77438656858, "data_wait_s": 1.0326000847271644e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:44136496bae15c1b5352966cc829ff2a992136d9c2c902dd52faab1009befaad
-size 317313
+oid sha256:a7661dcb872529b428ba53ae1347d2eb06260ace14496088b3df913d8cd24b40
+size 319431