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": 77036.49945384974,
-      "required_pages_per_second_upper_estimate": 0.005329563781522906
+      "required_token_positions_per_second": 76050.58972119447,
+      "required_pages_per_second_upper_estimate": 0.005261356258592062
     },
     {
       "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": 9629.562431731218,
-      "required_pages_per_second_upper_estimate": 0.004031696702035793
+      "required_token_positions_per_second": 9506.323715149309,
+      "required_pages_per_second_upper_estimate": 0.003980099224919994
     },
     {
       "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": 19259.124863462435,
-      "required_pages_per_second_upper_estimate": 0.004011167811781963
+      "required_token_positions_per_second": 19012.647430298617,
+      "required_pages_per_second_upper_estimate": 0.003959833062500961
     },
     {
       "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": 28888.687295193653,
-      "required_pages_per_second_upper_estimate": 0.0017318326928754389
+      "required_token_positions_per_second": 28518.971145447926,
+      "required_pages_per_second_upper_estimate": 0.0017096687742220562
     },
     {
       "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": 28888.687295193653,
-      "required_pages_per_second_upper_estimate": 0.0022459683692948452
+      "required_token_positions_per_second": 28518.971145447926,
+      "required_pages_per_second_upper_estimate": 0.0022172245648615944
     },
     {
       "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": 28888.687295193653,
-      "required_pages_per_second_upper_estimate": 0.3705530623669996
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28518.971145447926,
+      "required_pages_per_second_upper_estimate": 0.3658107405683345
     }
   ],
   "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": 0.09568454871865129,
-    "producer_p95_seconds": 0.22755040199990617,
-    "gpu_update_seconds": 0.957052832393666,
-    "producer_batches_per_second": 10.451008165805094,
-    "gpu_batches_per_second": 1.0448743958041686,
-    "producer_headroom": 10.002166966453098,
+    "producer_batch_seconds": 0.385939680726608,
+    "producer_p95_seconds": 0.23219145200346247,
+    "gpu_update_seconds": 0.9694599380529565,
+    "producer_batches_per_second": 2.591078476608836,
+    "gpu_batches_per_second": 1.0315021392306107,
+    "producer_headroom": 2.511946779423551,
     "sustainable": true,
-    "buffer_seconds": 60.29432844080096,
+    "buffer_seconds": 60.1065161592833,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397936.0304517304,
-    "upload_bytes_per_second": 5266067.482656801,
+    "generation_bytes_per_second": 357731.9527391862,
+    "upload_bytes_per_second": 5262673.944236853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.3878086582559,
+    "estimated_full_shards_per_hour": 76.76094948417366,
     "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": 1.0760069800016936,
-      "bytes": 5947734,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2793344790006813,
       "bytes": 12370460,
@@ -562,6 +556,12 @@
       "bytes": 11531600,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.4267704899975797,
+      "bytes": 7513990,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -193,3 +193,4 @@
 {"step": 623700, "loss": 0.5462983846664429, "lr": 0.0001, "grad_norm": 0.17555965936058146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91293035014, "tokens_per_second": 151753.54044312995, "data_wait_s": 1.3606000720756128e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 623800, "loss": 0.5751963257789612, "lr": 0.0001, "grad_norm": 0.15222753362728927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91307770224, "tokens_per_second": 152999.87859197438, "data_wait_s": 1.3946999388281256e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 623900, "loss": 0.55637127161026, "lr": 0.0001, "grad_norm": 0.17060754407015657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91322495048, "tokens_per_second": 152848.3170181749, "data_wait_s": 1.3158998626749963e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 624000, "loss": 0.5446088910102844, "lr": 0.0001, "grad_norm": 0.15555424520469768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91337146397, "tokens_per_second": 151456.037639442, "data_wait_s": 1.241100108018145e-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 sha256:3d8cbdd7257a1e75a6d2cd060ff31d678a5a0c3403db6de690f312ffc888a8ba
-size 416307
+oid sha256:7718f9e2924b892a549f2312389886ba0f6d927871924d9d2593a34af61351ed
+size 418425