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": 76745.35830844473,
-      "required_pages_per_second_upper_estimate": 0.005309421961543259
+      "required_token_positions_per_second": 77102.70327089421,
+      "required_pages_per_second_upper_estimate": 0.005334143915200062
     },
     {
       "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": 9593.169788555591,
-      "required_pages_per_second_upper_estimate": 0.004016459862302966
+      "required_token_positions_per_second": 9637.837908861777,
+      "required_pages_per_second_upper_estimate": 0.004035161471498744
     },
     {
       "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": 19186.339577111183,
-      "required_pages_per_second_upper_estimate": 0.003996008556112078
+      "required_token_positions_per_second": 19275.675817723553,
+      "required_pages_per_second_upper_estimate": 0.0040146149390765384
     },
     {
       "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": 28779.509365666774,
-      "required_pages_per_second_upper_estimate": 0.0017252876427053484
+      "required_token_positions_per_second": 28913.51372658533,
+      "required_pages_per_second_upper_estimate": 0.001733320999529604
     },
     {
       "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": 28779.509365666774,
-      "required_pages_per_second_upper_estimate": 0.002237480265497091
+      "required_token_positions_per_second": 28913.51372658533,
+      "required_pages_per_second_upper_estimate": 0.0022478985151356172
     },
     {
       "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": 28779.509365666774,
-      "required_pages_per_second_upper_estimate": 0.36915264511315626
+      "required_token_positions_per_second": 28913.51372658533,
+      "required_pages_per_second_upper_estimate": 0.3708715091723468
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2614797119843928,
-    "producer_p95_seconds": 0.22375611399911577,
-    "gpu_update_seconds": 0.9606835074465642,
-    "producer_batches_per_second": 3.824388486628316,
-    "gpu_batches_per_second": 1.0409255412929244,
-    "producer_headroom": 3.6740269451723484,
+    "producer_batch_seconds": 0.10579770080467199,
+    "producer_p95_seconds": 0.23710841899992374,
+    "gpu_update_seconds": 0.9562310641815313,
+    "producer_batches_per_second": 9.452001247609724,
+    "gpu_batches_per_second": 1.0457723425414254,
+    "producer_headroom": 9.038297211647008,
     "sustainable": true,
-    "buffer_seconds": 60.523060969133546,
+    "buffer_seconds": 60.24255704343647,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335486.6609085322,
-    "upload_bytes_per_second": 5916043.698961132,
+    "generation_bytes_per_second": 338193.1031924008,
+    "upload_bytes_per_second": 5929079.9720342085,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.98762770120597,
+    "estimated_full_shards_per_hour": 72.56836721257227,
     "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": 1.3033395869997548,
-      "bytes": 10049773,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.8627702979993046,
       "bytes": 2455057,
@@ -562,6 +556,12 @@
       "bytes": 11776139,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1994448559999,
+      "bytes": 10381245,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 497600, "loss": 0.5791856050491333, "lr": 0.0003, "grad_norm": 0.1502606423125578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72832209251, "tokens_per_second": 154244.70347143302, "data_wait_s": 1.1795998943853192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 497700, "loss": 0.5489344000816345, "lr": 0.0003, "grad_norm": 0.13772265310018428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72846682702, "tokens_per_second": 145832.15411135752, "data_wait_s": 1.3814000340062194e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 497800, "loss": 0.5586634278297424, "lr": 0.0003, "grad_norm": 0.17908025476657188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72861238788, "tokens_per_second": 152143.9408408102, "data_wait_s": 1.841000084823463e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 497900, "loss": 0.569886326789856, "lr": 0.0003, "grad_norm": 0.17307215126794653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72875927042, "tokens_per_second": 153519.26389961864, "data_wait_s": 1.4635001207352616e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:acfb3782c13b69c418cafb8353853ea698c32e25f2c0a69c611b904dd6b02bcc
-size 272835
+oid sha256:ca035d9eb87c3c0db6710bfccf40b4dfb0bf3505ab083b6371160223e8dfa96c
+size 274953