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": 77214.44896319561,
-      "required_pages_per_second_upper_estimate": 0.005341874741479202
+      "required_token_positions_per_second": 77199.20039764933,
+      "required_pages_per_second_upper_estimate": 0.0053408198103331635
     },
     {
       "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": 9651.806120399451,
-      "required_pages_per_second_upper_estimate": 0.0040410096699801425
+      "required_token_positions_per_second": 9649.900049706166,
+      "required_pages_per_second_upper_estimate": 0.0040402116379841466
     },
     {
       "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": 19303.612240798902,
-      "required_pages_per_second_upper_estimate": 0.004020433359270114
+      "required_token_positions_per_second": 19299.800099412332,
+      "required_pages_per_second_upper_estimate": 0.00401963939075222
     },
     {
       "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": 28955.418361198354,
-      "required_pages_per_second_upper_estimate": 0.0017358331183899827
+      "required_token_positions_per_second": 28949.700149118496,
+      "required_pages_per_second_upper_estimate": 0.0017354903203069996
     },
     {
       "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": 28955.418361198354,
-      "required_pages_per_second_upper_estimate": 0.0022511564161577713
+      "required_token_positions_per_second": 28949.700149118496,
+      "required_pages_per_second_upper_estimate": 0.0022507118503203192
     },
     {
       "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": 28955.418361198354,
-      "required_pages_per_second_upper_estimate": 0.37140901683147154
+      "required_token_positions_per_second": 28949.700149118496,
+      "required_pages_per_second_upper_estimate": 0.37133566974664894
     }
   ],
   "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": 0.09311686698426058,
-    "producer_p95_seconds": 0.22002124699974956,
-    "gpu_update_seconds": 0.9548471949225794,
-    "producer_batches_per_second": 10.739192934498416,
-    "gpu_batches_per_second": 1.047287990494732,
-    "producer_headroom": 10.254288249238195,
+    "producer_batch_seconds": 0.11470474844537648,
+    "producer_p95_seconds": 0.23065813799985335,
+    "gpu_update_seconds": 0.9550357985604858,
+    "producer_batches_per_second": 8.718034898757567,
+    "gpu_batches_per_second": 1.0470811685879087,
+    "producer_headroom": 8.326035421413117,
     "sustainable": true,
-    "buffer_seconds": 60.155373280122504,
+    "buffer_seconds": 60.167255309310605,
     "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": 422974.3690253121,
-    "upload_bytes_per_second": 5889673.536274945,
+    "generation_bytes_per_second": 424303.18634716957,
+    "upload_bytes_per_second": 5785623.808500083,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.76045325345537,
+    "estimated_full_shards_per_hour": 91.04558651744189,
     "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": 1.3173817269998835,
-      "bytes": 11936639,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9754752660001031,
       "bytes": 2583341,
@@ -562,6 +556,12 @@
       "bytes": 6116642,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.95897385599892,
+      "bytes": 2279278,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -114,3 +114,4 @@
 {"step": 496400, "loss": 0.5905017852783203, "lr": 0.0003, "grad_norm": 0.15937914559476943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72656150286, "tokens_per_second": 151870.15509323424, "data_wait_s": 1.6418000086559914e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 496500, "loss": 0.6291759610176086, "lr": 0.0003, "grad_norm": 0.2717890904557022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72670760815, "tokens_per_second": 152419.8958035513, "data_wait_s": 1.2229000276420265e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 496600, "loss": 0.6377224326133728, "lr": 0.0003, "grad_norm": 0.1596387510088863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72685404914, "tokens_per_second": 152688.65239828298, "data_wait_s": 1.4692999684484676e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 496700, "loss": 0.5884827375411987, "lr": 0.0003, "grad_norm": 0.17035361001570282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72700157012, "tokens_per_second": 154287.04699583387, "data_wait_s": 1.2817001334042288e-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 sha256:0515e94a884ba7ee4b2bb0c4d026e1140e504b39c9ab79762e2c8ceb0b79b33a
-size 247419
+oid sha256:96a2852bc0d92bfd6cc86cf285aced1ebe7db97ddae2bb1bbad28243185d62e9
+size 249537