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": 77077.35602287696,
-      "required_pages_per_second_upper_estimate": 0.0053323903337685104
+      "required_token_positions_per_second": 76865.29907421028,
+      "required_pages_per_second_upper_estimate": 0.005317719742019839
     },
     {
       "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": 9634.66950285962,
-      "required_pages_per_second_upper_estimate": 0.004033834926069482
+      "required_token_positions_per_second": 9608.162384276286,
+      "required_pages_per_second_upper_estimate": 0.00402273694905022
     },
     {
       "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": 19269.33900571924,
-      "required_pages_per_second_upper_estimate": 0.00401329514824901
+      "required_token_positions_per_second": 19216.32476855257,
+      "required_pages_per_second_upper_estimate": 0.004002253680726644
     },
     {
       "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": 28904.008508578863,
-      "required_pages_per_second_upper_estimate": 0.001732751176223744
+      "required_token_positions_per_second": 28824.487152828857,
+      "required_pages_per_second_upper_estimate": 0.0017279839923686107
     },
     {
       "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": 28904.008508578863,
-      "required_pages_per_second_upper_estimate": 0.0022471595262446497
+      "required_token_positions_per_second": 28824.487152828857,
+      "required_pages_per_second_upper_estimate": 0.002240977090612565
     },
     {
       "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": 28904.008508578863,
-      "required_pages_per_second_upper_estimate": 0.3707495864416678
+      "required_token_positions_per_second": 28824.487152828857,
+      "required_pages_per_second_upper_estimate": 0.3697295718734862
     }
   ],
   "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": 0.07764825636709816,
-    "producer_p95_seconds": 0.218513736002933,
-    "gpu_update_seconds": 0.9565455252268531,
-    "producer_batches_per_second": 12.87858925347008,
-    "gpu_batches_per_second": 1.045428548487372,
-    "producer_headroom": 12.318956921641444,
+    "producer_batch_seconds": 0.10964426420312634,
+    "producer_p95_seconds": 0.2337341689999448,
+    "gpu_update_seconds": 0.959184454988182,
+    "producer_batches_per_second": 9.12040412937065,
+    "gpu_batches_per_second": 1.0425523420438678,
+    "producer_headroom": 8.74814986410235,
     "sustainable": true,
-    "buffer_seconds": 60.26236808929175,
+    "buffer_seconds": 60.42862066425546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540510.3690120778,
-    "upload_bytes_per_second": 5146835.463529797,
+    "generation_bytes_per_second": 551156.8628531881,
+    "upload_bytes_per_second": 5155933.0149229895,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.98094275256872,
+    "estimated_full_shards_per_hour": 118.26543249317866,
     "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": 1.0021623419997923,
-      "bytes": 6273488,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0131233649990463,
       "bytes": 2204351,
@@ -562,6 +556,12 @@
       "bytes": 6090581,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1632558510027593,
+      "bytes": 7851564,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 577500, "loss": 0.574500560760498, "lr": 0.0003, "grad_norm": 0.1523363075603015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84528550416, "tokens_per_second": 152968.316301744, "data_wait_s": 8.859002264216542e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 577600, "loss": 0.5894696116447449, "lr": 0.0003, "grad_norm": 0.1705478396303194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84543370450, "tokens_per_second": 154447.42466988543, "data_wait_s": 1.2088999937986955e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 577700, "loss": 0.5548363924026489, "lr": 0.0003, "grad_norm": 0.16746688977543625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84558062867, "tokens_per_second": 153351.48063650445, "data_wait_s": 8.284001523861662e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 577800, "loss": 0.5663211941719055, "lr": 0.0003, "grad_norm": 0.19217156588652878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84572541026, "tokens_per_second": 150382.1330178825, "data_wait_s": 1.0037998436018825e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:feb97729dd6e540dcdf247fdc2fa6d010fbc977e12d2b23e66ba0dfbd5fda8c3
-size 439605
+oid sha256:779d5677bbb875ca4f538257197d20a5ced577304e691175d20d70ee439dda17
+size 441723