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": 76554.8148855621,
-      "required_pages_per_second_upper_estimate": 0.005296239725426585
+      "required_token_positions_per_second": 77031.49068797404,
+      "required_pages_per_second_upper_estimate": 0.005329217263477687
     },
     {
       "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": 9569.351860695262,
-      "required_pages_per_second_upper_estimate": 0.0040064877932827385
+      "required_token_positions_per_second": 9628.936335996756,
+      "required_pages_per_second_upper_estimate": 0.004031434568826139
     },
     {
       "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": 19138.703721390524,
-      "required_pages_per_second_upper_estimate": 0.003986087263607458
+      "required_token_positions_per_second": 19257.87267199351,
+      "required_pages_per_second_upper_estimate": 0.004010907013321496
     },
     {
       "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": 28708.055582085788,
-      "required_pages_per_second_upper_estimate": 0.0017210040974832785
+      "required_token_positions_per_second": 28886.809007990265,
+      "required_pages_per_second_upper_estimate": 0.0017317200924255743
     },
     {
       "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": 28708.055582085788,
-      "required_pages_per_second_upper_estimate": 0.002231925048115648
+      "required_token_positions_per_second": 28886.809007990265,
+      "required_pages_per_second_upper_estimate": 0.002245822340726494
     },
     {
       "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": 28708.055582085788,
-      "required_pages_per_second_upper_estimate": 0.3682361126984747
+      "required_token_positions_per_second": 28886.809007990265,
+      "required_pages_per_second_upper_estimate": 0.3705289697155022
     }
   ],
   "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": 0.7132447240858255,
-    "producer_p95_seconds": 0.3223656980007945,
-    "gpu_update_seconds": 0.9630746297305041,
-    "producer_batches_per_second": 1.4020433186964154,
-    "gpu_batches_per_second": 1.0383411307178019,
-    "producer_headroom": 1.3502723500196774,
+    "producer_batch_seconds": 0.10171361467956785,
+    "producer_p95_seconds": 0.3008065889989666,
+    "gpu_update_seconds": 0.9571150621846947,
+    "producer_batches_per_second": 9.831525535203294,
+    "gpu_batches_per_second": 1.0448064600690923,
+    "producer_headroom": 9.409901173996515,
     "sustainable": true,
-    "buffer_seconds": 60.67370167302175,
+    "buffer_seconds": 60.29824891763576,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333256.8913321526,
-    "upload_bytes_per_second": 5391361.125655269,
+    "generation_bytes_per_second": 320066.6025585662,
+    "upload_bytes_per_second": 5515946.4320974825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.50917105649408,
+    "estimated_full_shards_per_hour": 68.6788421398901,
     "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": 1.0528518130013254,
-      "bytes": 5770251,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.3281948670010024,
       "bytes": 5971602,
@@ -562,6 +556,12 @@
       "bytes": 859128,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.4843631869989622,
+      "bytes": 18084349,
+      "pages": 33,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -295,3 +295,4 @@
 {"step": 469700, "loss": 0.6009871363639832, "lr": 0.0003, "grad_norm": 0.1570330550552947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68749833614, "tokens_per_second": 151178.1979620952, "data_wait_s": 9.02800093172118e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 469800, "loss": 0.5855575203895569, "lr": 0.0003, "grad_norm": 0.1524110247525948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68764429425, "tokens_per_second": 151972.78918288607, "data_wait_s": 1.3544002285925671e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 469900, "loss": 0.5873199701309204, "lr": 0.0003, "grad_norm": 0.16370151859587578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68779142018, "tokens_per_second": 130762.99191042948, "data_wait_s": 9.157000022241846e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 470000, "loss": 0.591810941696167, "lr": 0.0003, "grad_norm": 0.16283904999133825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68793746250, "tokens_per_second": 151549.12583840796, "data_wait_s": 1.3904002116760239e-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 sha256:6dd11c379235f0407aff6e8a8efed7b08656143ca8052f42f5e2dfeefea70b3f
-size 633126
+oid sha256:e8a25eb5dbe5e1f4dc5275826952d47cc17d6fdd4722f9a2ad27396be2653f03
+size 635244