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": 76227.96706083804,
-      "required_pages_per_second_upper_estimate": 0.005273627634520735
+      "required_token_positions_per_second": 75738.94746480303,
+      "required_pages_per_second_upper_estimate": 0.005239796124185242
     },
     {
       "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": 9528.495882604755,
-      "required_pages_per_second_upper_estimate": 0.003989382248426082
+      "required_token_positions_per_second": 9467.368433100379,
+      "required_pages_per_second_upper_estimate": 0.0039637894618049
     },
     {
       "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": 19056.99176520951,
-      "required_pages_per_second_upper_estimate": 0.003969068818024148
+      "required_token_positions_per_second": 18934.736866200758,
+      "required_pages_per_second_upper_estimate": 0.003943606346639122
     },
     {
       "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": 28585.48764781426,
-      "required_pages_per_second_upper_estimate": 0.0017136563369740997
+      "required_token_positions_per_second": 28402.105299301136,
+      "required_pages_per_second_upper_estimate": 0.001702662845189371
     },
     {
       "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": 28585.48764781426,
-      "required_pages_per_second_upper_estimate": 0.0022223959303454034
+      "required_token_positions_per_second": 28402.105299301136,
+      "required_pages_per_second_upper_estimate": 0.002208138759362214
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28585.48764781426,
-      "required_pages_per_second_upper_estimate": 0.36666394284083403
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28402.105299301136,
+      "required_pages_per_second_upper_estimate": 0.3643117109747327
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.437405773734298,
-    "producer_p95_seconds": 0.26152585000090767,
-    "gpu_update_seconds": 0.9672040701434056,
-    "producer_batches_per_second": 2.2862066759261612,
-    "gpu_batches_per_second": 1.0339079733729117,
-    "producer_headroom": 2.2112284021448088,
+    "producer_batch_seconds": 0.0886448418124246,
+    "producer_p95_seconds": 0.27419541999915964,
+    "gpu_update_seconds": 0.9734489647385508,
+    "producer_batches_per_second": 11.280972243326158,
+    "gpu_batches_per_second": 1.0272752206055098,
+    "producer_headroom": 10.981450751510176,
     "sustainable": true,
-    "buffer_seconds": 60.933856419034555,
+    "buffer_seconds": 60.35383581379015,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298903.073504746,
-    "upload_bytes_per_second": 5223106.882505676,
+    "generation_bytes_per_second": 303151.24014358345,
+    "upload_bytes_per_second": 5204216.251252829,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.1376414666823,
+    "estimated_full_shards_per_hour": 65.04919913511874,
     "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": 1.3655256990005,
-      "bytes": 10029580,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8411326719997305,
       "bytes": 2272552,
@@ -562,6 +556,12 @@
       "bytes": 6135652,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0582829949998995,
+      "bytes": 6869016,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 625300, "loss": 0.5687322616577148, "lr": 0.0001, "grad_norm": 0.1808495224529712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91527167819, "tokens_per_second": 152496.99468937732, "data_wait_s": 2.0665000192821026e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625400, "loss": 0.5248603224754333, "lr": 0.0001, "grad_norm": 0.1754910851709876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91541938960, "tokens_per_second": 153515.0365525391, "data_wait_s": 1.5395002264995128e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625500, "loss": 0.5510759353637695, "lr": 0.0001, "grad_norm": 0.16776120099390332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91556609596, "tokens_per_second": 151143.33589475485, "data_wait_s": 1.6912999853957444e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 625600, "loss": 0.5008061528205872, "lr": 0.0001, "grad_norm": 0.16755687114194404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91571384588, "tokens_per_second": 153110.7857187314, "data_wait_s": 1.8451002688379958e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7270c5f14121ed45a4499502f92846c662c0be005c07f2a97859ba868073bfb3
-size 450978
+oid sha256:2abcbd70682131586772a8d9fc46a8209c85207ba6aac3f50d6afe1fd501297d
+size 453096