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": 75360.5462369287,
-      "required_pages_per_second_upper_estimate": 0.006532061148258745
+      "required_token_positions_per_second": 74232.14774163831,
+      "required_pages_per_second_upper_estimate": 0.006434254426586814
     },
     {
       "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": 9420.068279616087,
-      "required_pages_per_second_upper_estimate": 0.0037578060793107097
+      "required_token_positions_per_second": 9279.01846770479,
+      "required_pages_per_second_upper_estimate": 0.0037015392004566736
     },
     {
       "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": 18840.136559232174,
-      "required_pages_per_second_upper_estimate": 0.004358334017899616
+      "required_token_positions_per_second": 18558.03693540958,
+      "required_pages_per_second_upper_estimate": 0.00429307523471207
     },
     {
       "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": 28260.20483884826,
-      "required_pages_per_second_upper_estimate": 0.0017208982048499664
+      "required_token_positions_per_second": 27837.05540311437,
+      "required_pages_per_second_upper_estimate": 0.0016951306243072838
     },
     {
       "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": 28260.20483884826,
-      "required_pages_per_second_upper_estimate": 0.0021755487977484283
+      "required_token_positions_per_second": 27837.05540311437,
+      "required_pages_per_second_upper_estimate": 0.002142973582833025
     },
     {
       "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": 28260.20483884826,
-      "required_pages_per_second_upper_estimate": 0.3508971632771057
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27837.05540311437,
+      "required_pages_per_second_upper_estimate": 0.34564306347535134
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08665672595301999,
-    "producer_p95_seconds": 0.2928277579994756,
-    "gpu_update_seconds": 0.9783368577001011,
-    "producer_batches_per_second": 11.539785158074622,
-    "gpu_batches_per_second": 1.022142825479176,
-    "producer_headroom": 11.289797150084992,
+    "producer_batch_seconds": 0.44042707221876753,
+    "producer_p95_seconds": 0.2901448610009538,
+    "gpu_update_seconds": 0.9932084985147812,
+    "producer_batches_per_second": 2.2705234602456117,
+    "gpu_batches_per_second": 1.0068379413742177,
+    "producer_headroom": 2.2551031967931294,
     "sustainable": true,
-    "buffer_seconds": 60.656885177406274,
+    "buffer_seconds": 60.58571840940166,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244758.29993424192,
-    "upload_bytes_per_second": 6469925.625094634,
+    "generation_bytes_per_second": 221302.37299426855,
+    "upload_bytes_per_second": 6449262.09604244,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.51943348427241,
+    "estimated_full_shards_per_hour": 47.48633758898775,
     "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": 1.3410022780008148,
-      "bytes": 8747781,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 0.8860283060002985,
       "bytes": 800987,
@@ -562,6 +556,12 @@
       "bytes": 2508350,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1389318259989523,
+      "bytes": 6045358,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -139,3 +139,4 @@
 {"step": 13900, "loss": 0.637822050601244, "lr": 0.0003, "grad_norm": 0.1495436076408611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2033595625, "tokens_per_second": 150288.13934782927, "data_wait_s": 1.3520000720745884e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 14000, "loss": 0.8031466826796532, "lr": 0.0003, "grad_norm": 0.2034141855442637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2048451112, "tokens_per_second": 152011.57959002908, "data_wait_s": 1.369399979012087e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 14100, "loss": 0.6691518537700176, "lr": 0.0003, "grad_norm": 0.15031498952481898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2063123218, "tokens_per_second": 149280.03400993772, "data_wait_s": 1.324299955740571e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 14200, "loss": 0.7174923568964005, "lr": 0.0003, "grad_norm": 0.14407498044637804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2077600356, "tokens_per_second": 147670.8799376657, "data_wait_s": 1.1600999641814269e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0a489bdd5e487f88e3ff725b031a9796afba80bd6855c18cad1d8070eef5a19
-size 295375
+oid sha256:63289d2ea312c4c1ed4b8e8c687428da4301f96f9fb12f4148e1c69f9ec867ea
+size 297458