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": 76061.44275743277,
-      "required_pages_per_second_upper_estimate": 0.005262107096821541
+      "required_token_positions_per_second": 76918.46836745046,
+      "required_pages_per_second_upper_estimate": 0.005321398117095952
     },
     {
       "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": 9507.680344679096,
-      "required_pages_per_second_upper_estimate": 0.00398066721737447
+      "required_token_positions_per_second": 9614.808545931308,
+      "required_pages_per_second_upper_estimate": 0.004025519558147541
     },
     {
       "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": 19015.360689358193,
-      "required_pages_per_second_upper_estimate": 0.003960398162809618
+      "required_token_positions_per_second": 19229.617091862616,
+      "required_pages_per_second_upper_estimate": 0.0040050221211299876
     },
     {
       "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": 28523.04103403729,
-      "required_pages_per_second_upper_estimate": 0.0017099127578286287
+      "required_token_positions_per_second": 28844.425637793927,
+      "required_pages_per_second_upper_estimate": 0.001729179274097963
     },
     {
       "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": 28523.04103403729,
-      "required_pages_per_second_upper_estimate": 0.0022175409807978696
+      "required_token_positions_per_second": 28844.425637793927,
+      "required_pages_per_second_upper_estimate": 0.0022425272201184723
     },
     {
       "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": 28523.04103403729,
-      "required_pages_per_second_upper_estimate": 0.36586294472925296
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28844.425637793927,
+      "required_pages_per_second_upper_estimate": 0.3699853213503409
     }
   ],
   "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": 0.09968073893756468,
-    "producer_p95_seconds": 0.2970630080017145,
-    "gpu_update_seconds": 0.969321607994285,
-    "producer_batches_per_second": 10.032028360326992,
-    "gpu_batches_per_second": 1.031649342955631,
-    "producer_headroom": 9.72426186167643,
+    "producer_batch_seconds": 0.20601662330460613,
+    "producer_p95_seconds": 0.2992978590009443,
+    "gpu_update_seconds": 0.9585214261910527,
+    "producer_batches_per_second": 4.853977237173957,
+    "gpu_batches_per_second": 1.0432734967373383,
+    "producer_headroom": 4.652641184074888,
     "sustainable": true,
-    "buffer_seconds": 60.09793969564567,
+    "buffer_seconds": 60.38684985003632,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 501234.5263153298,
-    "upload_bytes_per_second": 5004781.049397499,
+    "generation_bytes_per_second": 470261.9608047331,
+    "upload_bytes_per_second": 5035290.729280228,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.5532612046711,
+    "estimated_full_shards_per_hour": 100.90726965052123,
     "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": 1.1044216970003617,
-      "bytes": 7366106,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.023783693999576,
       "bytes": 5395680,
@@ -562,6 +556,12 @@
       "bytes": 2638070,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1678417209986947,
+      "bytes": 10292515,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 628600, "loss": 0.5532091856002808, "lr": 0.0001, "grad_norm": 0.15655342139928477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92010309733, "tokens_per_second": 150319.08527220646, "data_wait_s": 1.7439000657759607e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 628700, "loss": 0.49332788586616516, "lr": 0.0001, "grad_norm": 0.1571253458175584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92024910171, "tokens_per_second": 151198.03246507273, "data_wait_s": 2.006700015044771e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 628800, "loss": 0.5494332313537598, "lr": 0.0001, "grad_norm": 0.1732913685250206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92039581833, "tokens_per_second": 151701.541739374, "data_wait_s": 1.9781000446528196e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 628900, "loss": 0.5531455874443054, "lr": 0.0001, "grad_norm": 0.17502718310119103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92054220068, "tokens_per_second": 150524.7076351682, "data_wait_s": 1.763500040397048e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7745d01c30eaccafcb754ed02f9287856c0f0cbf76de21d5c56c371d40d65a99
-size 520872
+oid sha256:cd61dadd048efbe16262313386344dde7bb09dc614b5bdf0ec40e36b8120db0f
+size 522990