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": 76735.19107980974,
-      "required_pages_per_second_upper_estimate": 0.00530871856907507
+      "required_token_positions_per_second": 76721.56826703002,
+      "required_pages_per_second_upper_estimate": 0.005307776111277688
     },
     {
       "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": 9591.898884976217,
-      "required_pages_per_second_upper_estimate": 0.004015927761513767
+      "required_token_positions_per_second": 9590.196033378752,
+      "required_pages_per_second_upper_estimate": 0.0040152148131094925
     },
     {
       "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": 19183.797769952434,
-      "required_pages_per_second_upper_estimate": 0.003995479164712873
+      "required_token_positions_per_second": 19180.392066757504,
+      "required_pages_per_second_upper_estimate": 0.003994769846551802
     },
     {
       "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": 28775.69665492865,
-      "required_pages_per_second_upper_estimate": 0.0017250590765182702
+      "required_token_positions_per_second": 28770.588100136258,
+      "required_pages_per_second_upper_estimate": 0.0017247528264587777
     },
     {
       "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": 28775.69665492865,
-      "required_pages_per_second_upper_estimate": 0.002237183844007544
+      "required_token_positions_per_second": 28770.588100136258,
+      "required_pages_per_second_upper_estimate": 0.002236786676342593
     },
     {
       "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": 28775.69665492865,
-      "required_pages_per_second_upper_estimate": 0.36910373975357746
+      "required_token_positions_per_second": 28770.588100136258,
+      "required_pages_per_second_upper_estimate": 0.36903821269783943
     }
   ],
   "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": 0.22536556035169042,
-    "producer_p95_seconds": 0.3702471170036006,
-    "gpu_update_seconds": 0.9608107957054273,
-    "producer_batches_per_second": 4.437235212156937,
-    "gpu_batches_per_second": 1.0407876394288431,
-    "producer_headroom": 4.263343494924647,
+    "producer_batch_seconds": 0.5687725833749369,
+    "producer_p95_seconds": 0.35848651700507617,
+    "gpu_update_seconds": 0.9609813989123517,
+    "producer_batches_per_second": 1.7581719464505141,
+    "gpu_batches_per_second": 1.0406028682051598,
+    "producer_headroom": 1.6895705366284672,
     "sustainable": true,
-    "buffer_seconds": 60.53108012944192,
+    "buffer_seconds": 60.541828131478155,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340240.6376515004,
-    "upload_bytes_per_second": 4785228.511887805,
+    "generation_bytes_per_second": 286303.9963870822,
+    "upload_bytes_per_second": 4893644.916717925,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.00772044333229,
+    "estimated_full_shards_per_hour": 61.43417280873632,
     "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": 0.8264875540007779,
-      "bytes": 1223987,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.2477019600009953,
       "bytes": 12351474,
@@ -562,6 +556,12 @@
       "bytes": 838763,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.3304272509994917,
+      "bytes": 13303031,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -387,3 +387,4 @@
 {"step": 595800, "loss": 0.5828372240066528, "lr": 0.0003, "grad_norm": 0.16561907383262034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87207244523, "tokens_per_second": 153697.60219180965, "data_wait_s": 1.5946003259159625e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 595900, "loss": 0.539260745048523, "lr": 0.0003, "grad_norm": 0.16102720379060725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87222025481, "tokens_per_second": 153910.13551407494, "data_wait_s": 1.6432000847999007e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 596000, "loss": 0.5765054821968079, "lr": 0.0003, "grad_norm": 0.15422873717060467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87236448609, "tokens_per_second": 150704.33241506864, "data_wait_s": 1.3217999367043376e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 596100, "loss": 0.6018339991569519, "lr": 0.0003, "grad_norm": 0.17259046529465089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87251014095, "tokens_per_second": 137373.049811316, "data_wait_s": 1.0986004781443626e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a03a50db855dcc32b56ef7e15a6fef7c49a115b85fef2e3f941b514616ee716b
-size 830331
+oid sha256:215c07a8716daad39a04c1ab9f5082b6279bef0b1aa8d210d8c8b2d50dd3bb26
+size 832449