Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75739.11626551286,
-      "required_pages_per_second_upper_estimate": 0.006564874638859454
+      "required_token_positions_per_second": 74380.80905614789,
+      "required_pages_per_second_upper_estimate": 0.006447140012549847
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9467.389533189107,
-      "required_pages_per_second_upper_estimate": 0.003776683234876778
+      "required_token_positions_per_second": 9297.601132018486,
+      "required_pages_per_second_upper_estimate": 0.003708952103086998
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18934.779066378214,
-      "required_pages_per_second_upper_estimate": 0.004380227896276616
+      "required_token_positions_per_second": 18595.202264036972,
+      "required_pages_per_second_upper_estimate": 0.004301672779402573
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28402.16859956732,
-      "required_pages_per_second_upper_estimate": 0.0017295430530514678
+      "required_token_positions_per_second": 27892.80339605546,
+      "required_pages_per_second_upper_estimate": 0.001698525384590282
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28402.16859956732,
-      "required_pages_per_second_upper_estimate": 0.002186477561029423
+      "required_token_positions_per_second": 27892.80339605546,
+      "required_pages_per_second_upper_estimate": 0.0021472652176499516
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28402.16859956732,
-      "required_pages_per_second_upper_estimate": 0.352659878063093
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27892.80339605546,
+      "required_pages_per_second_upper_estimate": 0.3463352669711494
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12090066516398679,
-    "producer_p95_seconds": 0.2944767460012372,
-    "gpu_update_seconds": 0.9734467952007435,
-    "producer_batches_per_second": 8.271253087347565,
-    "gpu_batches_per_second": 1.0272775101116653,
-    "producer_headroom": 8.051624810172743,
+    "producer_batch_seconds": 0.0684249889295927,
+    "producer_p95_seconds": 0.22555203899901244,
+    "gpu_update_seconds": 0.991223420873856,
+    "producer_batches_per_second": 14.614543833232776,
+    "gpu_batches_per_second": 1.0088542894985337,
+    "producer_headroom": 14.486278132887909,
     "sustainable": true,
-    "buffer_seconds": 60.3537013024461,
+    "buffer_seconds": 60.464628673305214,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 489904.4653120648,
-    "upload_bytes_per_second": 4069243.3801506446,
+    "generation_bytes_per_second": 458525.58918752515,
+    "upload_bytes_per_second": 5832146.1743899975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.12209386369189,
+    "estimated_full_shards_per_hour": 98.38891751021687,
     "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": 32.774606008000774,
-      "bytes": 10411787,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.2439632269997674,
       "bytes": 7901436,
@@ -562,6 +556,12 @@
       "bytes": 2687904,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9390472680024686,
+      "bytes": 6039556,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -156,3 +156,4 @@
 {"step": 330600, "loss": 0.6046997979283333, "lr": 0.0003, "grad_norm": 0.19152793513634442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48390796240, "tokens_per_second": 149141.94140897636, "data_wait_s": 1.0780999218695797e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330700, "loss": 0.5738974809646606, "lr": 0.0003, "grad_norm": 0.1722429909213363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48405610354, "tokens_per_second": 151068.1783950212, "data_wait_s": 1.1067999366787262e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330800, "loss": 0.5701847933232784, "lr": 0.0003, "grad_norm": 0.15614280457714352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48420285230, "tokens_per_second": 147885.39125797685, "data_wait_s": 1.1126001481898129e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 330900, "loss": 0.6047540437430143, "lr": 0.0003, "grad_norm": 0.142320208239958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48434964339, "tokens_per_second": 149440.5948079516, "data_wait_s": 1.091599915525876e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b27bdee72c3554ce8b5a756ee9ea8cb4a67c1f37a1e2705d9edc064cb8a0847
-size 337158
+oid sha256:c722133d23481e1c01d5511c5788bceccf895fe324e8f9627b5921c34ff286f7
+size 339276