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": 76934.690366754,
-      "required_pages_per_second_upper_estimate": 0.005322520392647995
+      "required_token_positions_per_second": 76785.50545588862,
+      "required_pages_per_second_upper_estimate": 0.005312199434357644
     },
     {
       "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": 9616.83629584425,
-      "required_pages_per_second_upper_estimate": 0.0040263685346918566
+      "required_token_positions_per_second": 9598.188181986077,
+      "required_pages_per_second_upper_estimate": 0.0040185609588363385
     },
     {
       "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": 19233.6725916885,
-      "required_pages_per_second_upper_estimate": 0.004005866774792997
+      "required_token_positions_per_second": 19196.376363972155,
+      "required_pages_per_second_upper_estimate": 0.0039980989541273495
     },
     {
       "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": 28850.50888753275,
-      "required_pages_per_second_upper_estimate": 0.001729543955631221
+      "required_token_positions_per_second": 28794.56454595823,
+      "required_pages_per_second_upper_estimate": 0.001726190177775892
     },
     {
       "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": 28850.50888753275,
-      "required_pages_per_second_upper_estimate": 0.002243000166028278
+      "required_token_positions_per_second": 28794.56454595823,
+      "required_pages_per_second_upper_estimate": 0.002238650739544012
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28850.50888753275,
-      "required_pages_per_second_upper_estimate": 0.3700633507463058
+      "required_token_positions_per_second": 28794.56454595823,
+      "required_pages_per_second_upper_estimate": 0.3693457568009419
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2615879527343665,
-    "producer_p95_seconds": 0.16687891600031435,
-    "gpu_update_seconds": 0.9583193179634903,
-    "producer_batches_per_second": 3.8228060181940613,
-    "gpu_batches_per_second": 1.043493521684489,
-    "producer_headroom": 3.663468856062459,
+    "producer_batch_seconds": 0.10240917035151398,
+    "producer_p95_seconds": 0.17609337299927574,
+    "gpu_update_seconds": 0.9601812160024775,
+    "producer_batches_per_second": 9.764750525441752,
+    "gpu_batches_per_second": 1.0414700718300864,
+    "producer_headroom": 9.375930033479493,
     "sustainable": true,
-    "buffer_seconds": 60.37411703169989,
+    "buffer_seconds": 60.49141660815609,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440424.77772319113,
-    "upload_bytes_per_second": 4697119.616504709,
+    "generation_bytes_per_second": 455304.9333644603,
+    "upload_bytes_per_second": 4894354.335854951,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.50490473529625,
+    "estimated_full_shards_per_hour": 97.6978397436176,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 6226670,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.6309672779998436,
+      "bytes": 11574403,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -51,3 +51,4 @@
 {"step": 765100, "loss": 0.5478135943412781, "lr": 3e-05, "grad_norm": 0.24365099445058058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111988119969, "tokens_per_second": 141452.28978523877, "data_wait_s": 1.1377000191714615e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765200, "loss": 0.5334945321083069, "lr": 3e-05, "grad_norm": 0.25597445530657653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112002882488, "tokens_per_second": 152509.53592797738, "data_wait_s": 1.3847000445821323e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765300, "loss": 0.532582700252533, "lr": 3e-05, "grad_norm": 0.23488419984191805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112017439376, "tokens_per_second": 150846.18801539313, "data_wait_s": 1.6494999727001414e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 765400, "loss": 0.4953550100326538, "lr": 3e-05, "grad_norm": 0.2328708429836473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112031963356, "tokens_per_second": 150755.27137500362, "data_wait_s": 1.7504999959783163e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:472ec722a049c80ae00b8821b83af0a5fb8893787a8d31966fbfcb72c98787e6
-size 113202
+oid sha256:107b7c7e5cff7cabc09efe8619c538aa70bfd7315d9669164c91a6807ee47356
+size 115320