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": 76071.2503009277,
-      "required_pages_per_second_upper_estimate": 0.005262785605437156
+      "required_token_positions_per_second": 75160.00660651727,
+      "required_pages_per_second_upper_estimate": 0.005199743652281167
     },
     {
       "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": 9508.906287615962,
-      "required_pages_per_second_upper_estimate": 0.003981180494081536
+      "required_token_positions_per_second": 9395.000825814659,
+      "required_pages_per_second_upper_estimate": 0.003933490655841854
     },
     {
       "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": 19017.812575231925,
-      "required_pages_per_second_upper_estimate": 0.003960908825976538
+      "required_token_positions_per_second": 18790.001651629318,
+      "required_pages_per_second_upper_estimate": 0.003913461818368175
     },
     {
       "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": 28526.71886284789,
-      "required_pages_per_second_upper_estimate": 0.0017101332380501046
+      "required_token_positions_per_second": 28185.002477443973,
+      "required_pages_per_second_upper_estimate": 0.0016896478625158487
     },
     {
       "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": 28526.71886284789,
-      "required_pages_per_second_upper_estimate": 0.0022178269158108347
+      "required_token_positions_per_second": 28185.002477443973,
+      "required_pages_per_second_upper_estimate": 0.002191259969897212
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28526.71886284789,
-      "required_pages_per_second_upper_estimate": 0.3659101199682904
+      "required_token_positions_per_second": 28185.002477443973,
+      "required_pages_per_second_upper_estimate": 0.3615269490827975
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0928377404922287,
-    "producer_p95_seconds": 0.2791655979999632,
-    "gpu_update_seconds": 0.9691966374726573,
-    "producer_batches_per_second": 10.771481454610674,
-    "gpu_batches_per_second": 1.0317823662777739,
-    "producer_headroom": 10.439683606407753,
+    "producer_batch_seconds": 0.2782707595936529,
+    "producer_p95_seconds": 0.29377643600309966,
+    "gpu_update_seconds": 0.9809472261755616,
+    "producer_batches_per_second": 3.5936222744360853,
+    "gpu_batches_per_second": 1.0194228326621808,
+    "producer_headroom": 3.5251538020307907,
     "sustainable": true,
-    "buffer_seconds": 60.09019152330475,
+    "buffer_seconds": 60.81872802288482,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 516947.29165763385,
-    "upload_bytes_per_second": 3284054.4322956875,
+    "generation_bytes_per_second": 463119.66016296175,
+    "upload_bytes_per_second": 3239888.804448443,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.92485487267267,
+    "estimated_full_shards_per_hour": 99.37469819704665,
     "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": 1.8976086379989283,
-      "bytes": 13629725,
-      "pages": 49,
-      "shards": 4
-    },
     {
       "seconds": 1.6741066590002447,
       "bytes": 5692006,
@@ -562,6 +556,12 @@
       "bytes": 1206249,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 2.568047996999667,
+      "bytes": 10001629,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -260,3 +260,4 @@
 {"step": 826000, "loss": 0.5335186719894409, "lr": 3e-05, "grad_norm": 0.27717066949513663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120903085084, "tokens_per_second": 151215.27428752833, "data_wait_s": 9.088002116186544e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 826100, "loss": 0.49807727336883545, "lr": 3e-05, "grad_norm": 0.2930168926589278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120917748156, "tokens_per_second": 151171.80290622264, "data_wait_s": 1.2401997082633898e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 826200, "loss": 0.5254029035568237, "lr": 3e-05, "grad_norm": 0.27069934149591274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120932368378, "tokens_per_second": 150759.84045175835, "data_wait_s": 7.68699828768149e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 826300, "loss": 0.5496779084205627, "lr": 3e-05, "grad_norm": 0.2700586435145031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120947058412, "tokens_per_second": 152143.9024298574, "data_wait_s": 9.139999747276306e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a68f4458a455df277285fb810c1afa7686dcfdb6d10a752ac44872cf8c4ce104
-size 558996
+oid sha256:709a4e8c9bc9c6b82b1257437c3cb1aab4e4c26912915d58ac2c0255b7b462e1
+size 561114