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": 77502.5584724377,
-      "required_pages_per_second_upper_estimate": 0.005361806825834733
+      "required_token_positions_per_second": 76629.19223877549,
+      "required_pages_per_second_upper_estimate": 0.005301385323301123
     },
     {
       "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": 9687.819809054712,
-      "required_pages_per_second_upper_estimate": 0.004056087849368763
+      "required_token_positions_per_second": 9578.649029846936,
+      "required_pages_per_second_upper_estimate": 0.0040103803238595755
     },
     {
       "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": 19375.639618109424,
-      "required_pages_per_second_upper_estimate": 0.004035434762474221
+      "required_token_positions_per_second": 19157.298059693872,
+      "required_pages_per_second_upper_estimate": 0.003989959973910369
     },
     {
       "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": 29063.459427164136,
-      "required_pages_per_second_upper_estimate": 0.0017423100153254757
+      "required_token_positions_per_second": 28735.94708954081,
+      "required_pages_per_second_upper_estimate": 0.001722676150767341
     },
     {
       "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": 29063.459427164136,
-      "required_pages_per_second_upper_estimate": 0.002259556133814185
+      "required_token_positions_per_second": 28735.94708954081,
+      "required_pages_per_second_upper_estimate": 0.0022340934901384996
     },
     {
       "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": 29063.459427164136,
-      "required_pages_per_second_upper_estimate": 0.37279485162022213
+      "required_token_positions_per_second": 28735.94708954081,
+      "required_pages_per_second_upper_estimate": 0.3685938750085403
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0981617993592181,
-    "producer_p95_seconds": 0.23824944100124412,
-    "gpu_update_seconds": 0.9512976274998709,
-    "producer_batches_per_second": 10.187262321267676,
-    "gpu_batches_per_second": 1.051195725808888,
-    "producer_headroom": 9.691118476940767,
+    "producer_batch_seconds": 0.08593368737504647,
+    "producer_p95_seconds": 0.2581082110009447,
+    "gpu_update_seconds": 0.9621398561825444,
+    "producer_batches_per_second": 11.636879907592341,
+    "gpu_batches_per_second": 1.0393499381344333,
+    "producer_headroom": 11.196305960704436,
     "sustainable": true,
-    "buffer_seconds": 60.88304815999174,
+    "buffer_seconds": 60.614810939500295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268810.1422620847,
-    "upload_bytes_per_second": 4954596.749161316,
+    "generation_bytes_per_second": 249661.6029177761,
+    "upload_bytes_per_second": 4989090.300975377,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.6803989495936,
+    "estimated_full_shards_per_hour": 53.57156816148722,
     "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": 0.8532074099985039,
-      "bytes": 2480365,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 0.959426125000391,
       "bytes": 5985197,
@@ -562,6 +556,12 @@
       "bytes": 12054418,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 2.080120522998186,
+      "bytes": 11568040,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 461400, "loss": 0.5921451449394226, "lr": 0.0003, "grad_norm": 0.15366019789936122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67535009988, "tokens_per_second": 152302.3197972847, "data_wait_s": 1.3390999811235815e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 461500, "loss": 0.62558513879776, "lr": 0.0003, "grad_norm": 0.16961965440790405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67549806988, "tokens_per_second": 151838.21216204233, "data_wait_s": 1.6812999092508107e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 461600, "loss": 0.6056005954742432, "lr": 0.0003, "grad_norm": 0.16370604241757553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67564497122, "tokens_per_second": 130024.82292554397, "data_wait_s": 2.348600173718296e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 461700, "loss": 0.5677300691604614, "lr": 0.0003, "grad_norm": 0.1457644295961157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67579202027, "tokens_per_second": 152277.26054054042, "data_wait_s": 1.5961002645781264e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8f5c0251561df626cc6efb18e500677734282d6c6e9b9521a2994341b823b9a
-size 456549
+oid sha256:900ec56695642d7c8ee8f56d9c4a549fb43e8d99c72e5b8d761ea5f89a75ae2b
+size 458667