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": 74660.25095036415,
-      "required_pages_per_second_upper_estimate": 0.0051651693963956115
+      "required_token_positions_per_second": 76721.34780743423,
+      "required_pages_per_second_upper_estimate": 0.0053077608593713145
     },
     {
       "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": 9332.531368795519,
-      "required_pages_per_second_upper_estimate": 0.003907335998698544
+      "required_token_positions_per_second": 9590.168475929278,
+      "required_pages_per_second_upper_estimate": 0.004015203275380863
     },
     {
       "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": 18665.062737591037,
-      "required_pages_per_second_upper_estimate": 0.003887440337434735
+      "required_token_positions_per_second": 19180.336951858557,
+      "required_pages_per_second_upper_estimate": 0.003994758367571829
     },
     {
       "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": 27997.594106386554,
-      "required_pages_per_second_upper_estimate": 0.00167841301682697
+      "required_token_positions_per_second": 28770.505427787833,
+      "required_pages_per_second_upper_estimate": 0.001724747870377723
     },
     {
       "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": 27997.594106386554,
-      "required_pages_per_second_upper_estimate": 0.002176689793369805
+      "required_token_positions_per_second": 28770.505427787833,
+      "required_pages_per_second_upper_estimate": 0.0022367802489311407
     },
     {
       "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": 27997.594106386554,
-      "required_pages_per_second_upper_estimate": 0.3591230757223042
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28770.505427787833,
+      "required_pages_per_second_upper_estimate": 0.3690371522657205
     }
   ],
   "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": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12571840718754856,
+    "producer_batch_seconds": 0.3351845816563923,
     "producer_p95_seconds": 0.262184718001663,
-    "gpu_update_seconds": 0.9875134232942249,
-    "producer_batches_per_second": 7.954284677725715,
-    "gpu_batches_per_second": 1.0126444627599303,
-    "producer_headroom": 7.8549628919577215,
+    "gpu_update_seconds": 0.9609841603024579,
+    "producer_batches_per_second": 2.983430786279811,
+    "gpu_batches_per_second": 1.0405998780305206,
+    "producer_headroom": 2.867029728973606,
     "sustainable": true,
-    "buffer_seconds": 60.23831882094772,
+    "buffer_seconds": 60.54200209905485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331138.89534910914,
-    "upload_bytes_per_second": 4873605.143989582,
+    "generation_bytes_per_second": 308597.06108511257,
+    "upload_bytes_per_second": 4937549.809012917,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.05469842295604,
+    "estimated_full_shards_per_hour": 66.21774553694756,
     "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": 1.1292916710008285,
-      "bytes": 2686677,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3082805390004069,
       "bytes": 10490550,
@@ -562,6 +556,12 @@
       "bytes": 923291,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.0239220590010518,
+      "bytes": 7655031,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -256,3 +256,4 @@
 {"step": 785600, "loss": 0.5228918790817261, "lr": 3e-05, "grad_norm": 0.2482094230685791, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114989536499, "tokens_per_second": 151982.92673010292, "data_wait_s": 7.004000508459285e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 785700, "loss": 0.532866358757019, "lr": 3e-05, "grad_norm": 0.2698860448119994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115004283662, "tokens_per_second": 153080.66404407044, "data_wait_s": 1.0274998203385621e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 785800, "loss": 0.4968486726284027, "lr": 3e-05, "grad_norm": 0.25902219601183707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115018910117, "tokens_per_second": 152449.77224550906, "data_wait_s": 9.759001841302961e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 785900, "loss": 0.5158545970916748, "lr": 3e-05, "grad_norm": 0.26559567817473356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115033482450, "tokens_per_second": 150697.70523083094, "data_wait_s": 8.023998816497624e-06, "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:e2d5efd37d31bf654a1ca7858275831ba09786cfea2e9616772e2f0822d7df9d
-size 550524
+oid sha256:403c475bad047d29564d45663a58f20eef45554174eabefbdf7f3136ed594ba4
+size 552642