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": 76626.44251783224,
-      "required_pages_per_second_upper_estimate": 0.005301195091226038
+      "required_token_positions_per_second": 76994.47146261457,
+      "required_pages_per_second_upper_estimate": 0.005326656187570882
     },
     {
       "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": 9578.30531472903,
-      "required_pages_per_second_upper_estimate": 0.004010236417517295
+      "required_token_positions_per_second": 9624.308932826822,
+      "required_pages_per_second_upper_estimate": 0.004029497171750054
     },
     {
       "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": 19156.61062945806,
-      "required_pages_per_second_upper_estimate": 0.003989816800321004
+      "required_token_positions_per_second": 19248.617865653643,
+      "required_pages_per_second_upper_estimate": 0.004008979481226557
     },
     {
       "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": 28734.91594418709,
-      "required_pages_per_second_upper_estimate": 0.0017226143351778533
+      "required_token_positions_per_second": 28872.926798480465,
+      "required_pages_per_second_upper_estimate": 0.0017308878751623683
     },
     {
       "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": 28734.91594418709,
-      "required_pages_per_second_upper_estimate": 0.002234013323122777
+      "required_token_positions_per_second": 28872.926798480465,
+      "required_pages_per_second_upper_estimate": 0.0022447430599984936
     },
     {
       "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": 28734.91594418709,
-      "required_pages_per_second_upper_estimate": 0.3685806485831004
+      "required_token_positions_per_second": 28872.926798480465,
+      "required_pages_per_second_upper_estimate": 0.3703509036374657
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09030982552332034,
-    "producer_p95_seconds": 0.21598285999789368,
-    "gpu_update_seconds": 0.9621743823333867,
-    "producer_batches_per_second": 11.072992270832968,
-    "gpu_batches_per_second": 1.03931264265723,
-    "producer_headroom": 10.654149498771076,
+    "producer_batch_seconds": 0.3767553764296565,
+    "producer_p95_seconds": 0.21884131499973591,
+    "gpu_update_seconds": 0.9575752466305241,
+    "producer_batches_per_second": 2.6542421490478945,
+    "gpu_batches_per_second": 1.0443043546904103,
+    "producer_headroom": 2.54163658049167,
     "sustainable": true,
-    "buffer_seconds": 60.61698608700336,
+    "buffer_seconds": 60.327240537723014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416660.44699630915,
-    "upload_bytes_per_second": 3704325.9584100177,
+    "generation_bytes_per_second": 373556.7056534131,
+    "upload_bytes_per_second": 3703354.0885011475,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.40563256661373,
+    "estimated_full_shards_per_hour": 80.15657307817263,
     "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": 1.0782337069995265,
-      "bytes": 6047629,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.1856751459999941,
       "bytes": 11039128,
@@ -562,6 +556,12 @@
       "bytes": 6032843,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2020050520004588,
+      "bytes": 6391888,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -172,3 +172,4 @@
 {"step": 942200, "loss": 0.5340875387191772, "lr": 3e-05, "grad_norm": 0.3456524713919745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137911204235, "tokens_per_second": 152384.73326796383, "data_wait_s": 1.0294999810867012e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942300, "loss": 0.5293458700180054, "lr": 3e-05, "grad_norm": 0.34303656299887136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137925881125, "tokens_per_second": 153226.535566637, "data_wait_s": 9.727002179715782e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 942400, "loss": 0.520447313785553, "lr": 3e-05, "grad_norm": 0.3352271217179194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137940681345, "tokens_per_second": 146237.1304412144, "data_wait_s": 1.642300048843026e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 942500, "loss": 0.5312053561210632, "lr": 3e-05, "grad_norm": 0.3516962228627744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137955413024, "tokens_per_second": 153717.4198919749, "data_wait_s": 1.5871002688072622e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f2a9c14083cd8b127821c13e01c2f03cc96cc4bf4f21856c47435d1daff4214
-size 371046
+oid sha256:77cb80ff6d1b69bc84a96f71791858aee617b94ea3e9c733172551c96ff6454d
+size 373164