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": 74866.30083377541,
-      "required_pages_per_second_upper_estimate": 0.00648922120936694
+      "required_token_positions_per_second": 75329.11845854671,
+      "required_pages_per_second_upper_estimate": 0.00652933706808157
     },
     {
       "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": 9358.287604221927,
-      "required_pages_per_second_upper_estimate": 0.003733160844192567
+      "required_token_positions_per_second": 9416.139807318339,
+      "required_pages_per_second_upper_estimate": 0.003756238952975243
     },
     {
       "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": 18716.575208443854,
-      "required_pages_per_second_upper_estimate": 0.00432975027400024
+      "required_token_positions_per_second": 18832.279614636678,
+      "required_pages_per_second_upper_estimate": 0.0043565164520449505
     },
     {
       "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": 28074.86281266578,
-      "required_pages_per_second_upper_estimate": 0.0017096118478699125
+      "required_token_positions_per_second": 28248.41942195502,
+      "required_pages_per_second_upper_estimate": 0.0017201805347944717
     },
     {
       "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": 28074.86281266578,
-      "required_pages_per_second_upper_estimate": 0.002161280655513335
+      "required_token_positions_per_second": 28248.41942195502,
+      "required_pages_per_second_upper_estimate": 0.0021746415237318645
     },
     {
       "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": 28074.86281266578,
-      "required_pages_per_second_upper_estimate": 0.34859583561177815
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28248.41942195502,
+      "required_pages_per_second_upper_estimate": 0.35075082784254463
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.32341157330480996,
-    "producer_p95_seconds": 0.30776679900009185,
-    "gpu_update_seconds": 0.9847955512547258,
-    "producer_batches_per_second": 3.0920352966389264,
-    "gpu_batches_per_second": 1.0154391931664417,
-    "producer_headroom": 3.045022604452601,
+    "producer_batch_seconds": 0.09673103116418247,
+    "producer_p95_seconds": 0.21955628000068828,
+    "gpu_update_seconds": 0.9787450259433501,
+    "producer_batches_per_second": 10.337944173289033,
+    "gpu_batches_per_second": 1.0217165589538129,
+    "producer_headroom": 10.11821143808668,
     "sustainable": true,
-    "buffer_seconds": 60.07252862653828,
+    "buffer_seconds": 60.682191608487706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325906.4351306959,
-    "upload_bytes_per_second": 5147096.2862418825,
+    "generation_bytes_per_second": 326111.9211780653,
+    "upload_bytes_per_second": 5160657.26484681,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.93193426552446,
+    "estimated_full_shards_per_hour": 69.97602678781956,
     "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": 1.2236139670003467,
-      "bytes": 7766768,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.4731977419996838,
       "bytes": 6241189,
@@ -562,6 +556,12 @@
       "bytes": 6140094,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1522124919974885,
+      "bytes": 8515706,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -168,3 +168,4 @@
 {"step": 171800, "loss": 0.589970288798213, "lr": 0.0003, "grad_norm": 0.17102003820212547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25144044313, "tokens_per_second": 149047.31067176312, "data_wait_s": 1.2485001207096502e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 171900, "loss": 0.6244958490133286, "lr": 0.0003, "grad_norm": 0.14701528131234995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25158621255, "tokens_per_second": 148934.09620441706, "data_wait_s": 1.2003001756966114e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172000, "loss": 0.6820664405822754, "lr": 0.0003, "grad_norm": 0.14046266448044453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25173218574, "tokens_per_second": 149847.75509250868, "data_wait_s": 1.0988002031808719e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 172100, "loss": 0.5721213407814503, "lr": 0.0003, "grad_norm": 0.14904676025056426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25187867206, "tokens_per_second": 149166.60639444797, "data_wait_s": 9.974999557016417e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6497d095cfcad0fafbda343ad6352a354256f79efe6916e52bd16718fdafdd9e
-size 362574
+oid sha256:9d3327a0efb2e5cb0c0daa57fecb64c95b40554fdbb14ceca215b63a4ec24613
+size 364692