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": 75193.30943556807,
-      "required_pages_per_second_upper_estimate": 0.0065175654861746035
+      "required_token_positions_per_second": 73836.48329009506,
+      "required_pages_per_second_upper_estimate": 0.006399959234729429
     },
     {
       "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": 9399.163679446008,
-      "required_pages_per_second_upper_estimate": 0.00374946692175124
+      "required_token_positions_per_second": 9229.560411261882,
+      "required_pages_per_second_upper_estimate": 0.0036818096422777573
     },
     {
       "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": 18798.327358892016,
-      "required_pages_per_second_upper_estimate": 0.004348662195217716
+      "required_token_positions_per_second": 18459.120822523764,
+      "required_pages_per_second_upper_estimate": 0.004270192732860065
     },
     {
       "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": 28197.491038338027,
-      "required_pages_per_second_upper_estimate": 0.0017170792634327752
+      "required_token_positions_per_second": 27688.681233785646,
+      "required_pages_per_second_upper_estimate": 0.0016860954158542684
     },
     {
       "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": 28197.491038338027,
-      "required_pages_per_second_upper_estimate": 0.0021707209157822392
+      "required_token_positions_per_second": 27688.681233785646,
+      "required_pages_per_second_upper_estimate": 0.0021315513285521115
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28197.491038338027,
-      "required_pages_per_second_upper_estimate": 0.35011846776435707
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27688.681233785646,
+      "required_pages_per_second_upper_estimate": 0.3438007528686895
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05311391964065848,
-    "producer_p95_seconds": 0.19084293899959448,
-    "gpu_update_seconds": 0.9805127684023048,
-    "producer_batches_per_second": 18.827456281997012,
-    "gpu_batches_per_second": 1.019874531189888,
-    "producer_headroom": 18.460561281034256,
+    "producer_batch_seconds": 0.10207321441409789,
+    "producer_p95_seconds": 0.20086097000057634,
+    "gpu_update_seconds": 0.9985307630420475,
+    "producer_batches_per_second": 9.796889475265555,
+    "gpu_batches_per_second": 1.0014713987914368,
+    "producer_headroom": 9.78249552317552,
     "sustainable": true,
-    "buffer_seconds": 60.7917916409429,
+    "buffer_seconds": 60.910376545564894,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343074.3327639885,
-    "upload_bytes_per_second": 6209244.335175436,
+    "generation_bytes_per_second": 365619.55410723115,
+    "upload_bytes_per_second": 6240453.956569004,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.61576544942609,
+    "estimated_full_shards_per_hour": 78.45344512379361,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 16188168,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1706134349997228,
+      "bytes": 7624358,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -58,3 +58,4 @@
 {"step": 120800, "loss": 0.5996634922921658, "lr": 0.0003, "grad_norm": 0.14145665261984264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17677613798, "tokens_per_second": 141081.09919191344, "data_wait_s": 1.4481000107480213e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120900, "loss": 0.6196673363447189, "lr": 0.0003, "grad_norm": 0.14226493327640644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17692295129, "tokens_per_second": 148894.11960489125, "data_wait_s": 1.4530000044032931e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 121000, "loss": 0.6049286965280771, "lr": 0.0003, "grad_norm": 0.1259416569654147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17706865890, "tokens_per_second": 148902.3610268338, "data_wait_s": 1.737300044624135e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 121100, "loss": 0.6064906008541584, "lr": 0.0003, "grad_norm": 0.13850603728557837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17721498777, "tokens_per_second": 148900.58524872438, "data_wait_s": 1.1443999937910121e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e37913b361f7e6eb1bfe0e1d0685fd95a189ee68745585a5c7c2b0ac89bae07
-size 128028
+oid sha256:4c2ad2867fa03af4f058f8aca0020610dff4a5db6990cf082c6133be1bc65fd5
+size 130146