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": 75419.18551091442,
-      "required_pages_per_second_upper_estimate": 0.0065371438519079466
+      "required_token_positions_per_second": 74638.49930819964,
+      "required_pages_per_second_upper_estimate": 0.0064694759504877195
     },
     {
       "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": 9427.398188864303,
-      "required_pages_per_second_upper_estimate": 0.0037607300897017325
+      "required_token_positions_per_second": 9329.812413524955,
+      "required_pages_per_second_upper_estimate": 0.0037218016648815043
     },
     {
       "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": 18854.796377728606,
-      "required_pages_per_second_upper_estimate": 0.004361725308904772
+      "required_token_positions_per_second": 18659.62482704991,
+      "required_pages_per_second_upper_estimate": 0.004316575805557231
     },
     {
       "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": 28282.19456659291,
-      "required_pages_per_second_upper_estimate": 0.0017222372638984302
+      "required_token_positions_per_second": 27989.437240574865,
+      "required_pages_per_second_upper_estimate": 0.0017044098787229144
     },
     {
       "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": 28282.19456659291,
-      "required_pages_per_second_upper_estimate": 0.0021772416278616736
+      "required_token_positions_per_second": 27989.437240574865,
+      "required_pages_per_second_upper_estimate": 0.002154704358500658
     },
     {
       "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": 28282.19456659291,
-      "required_pages_per_second_upper_estimate": 0.3511702021008097
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27989.437240574865,
+      "required_pages_per_second_upper_estimate": 0.34753513590740737
     }
   ],
   "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": 0.12226313806280587,
-    "producer_p95_seconds": 0.38340045700169867,
-    "gpu_update_seconds": 0.9775761896729886,
-    "producer_batches_per_second": 8.1790801041464,
-    "gpu_batches_per_second": 1.0229381715347552,
-    "producer_headroom": 7.99567396324159,
+    "producer_batch_seconds": 0.28504399968772987,
+    "producer_p95_seconds": 0.39563816099689575,
+    "gpu_update_seconds": 0.9878012109482537,
+    "producer_batches_per_second": 3.5082303121466003,
+    "gpu_batches_per_second": 1.0123494372314406,
+    "producer_headroom": 3.465434150623782,
     "sustainable": true,
-    "buffer_seconds": 60.6097237597253,
+    "buffer_seconds": 60.255873867843476,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354070.7918018064,
-    "upload_bytes_per_second": 5958135.4166073045,
+    "generation_bytes_per_second": 366812.8488985609,
+    "upload_bytes_per_second": 6014367.581363513,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.97534957447667,
+    "estimated_full_shards_per_hour": 78.70949840753194,
     "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": 0.8470438740041573,
-      "bytes": 878239,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.8046116949990392,
       "bytes": 17147881,
@@ -562,6 +556,12 @@
       "bytes": 2510183,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0644101150028291,
+      "bytes": 6182736,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -322,3 +322,4 @@
 {"step": 117200, "loss": 0.6494732219725847, "lr": 0.0003, "grad_norm": 0.15238581215227845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17150398058, "tokens_per_second": 149314.99748354795, "data_wait_s": 1.3827004295308143e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 117300, "loss": 0.6032467819750309, "lr": 0.0003, "grad_norm": 0.1305913169480217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17165055303, "tokens_per_second": 149952.8754856903, "data_wait_s": 1.7099999240599573e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 117400, "loss": 0.5863179080188274, "lr": 0.0003, "grad_norm": 0.1443106834530696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17179656845, "tokens_per_second": 148130.04662776968, "data_wait_s": 1.2811004125978798e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 117500, "loss": 0.6004777606576681, "lr": 0.0003, "grad_norm": 0.13132205454677431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17194356680, "tokens_per_second": 149805.20721239422, "data_wait_s": 1.598800008650869e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4abe23a5280e37f519bf5539db4bf23d774737e67a3b6927c5645c723b05e903
-size 691095
+oid sha256:436e2c49372e856a6e35857519b8c73a3327abeabd178577dd83da203aaf637d
+size 693213