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": 77767.39282454367,
-      "required_pages_per_second_upper_estimate": 0.005380128680813783
+      "required_token_positions_per_second": 76049.66839064594,
+      "required_pages_per_second_upper_estimate": 0.005261292518806933
     },
     {
       "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": 9720.924103067959,
-      "required_pages_per_second_upper_estimate": 0.00406994792597584
+      "required_token_positions_per_second": 9506.208548830742,
+      "required_pages_per_second_upper_estimate": 0.003980051007187357
     },
     {
       "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": 19441.848206135917,
-      "required_pages_per_second_upper_estimate": 0.004049224265322257
+      "required_token_positions_per_second": 19012.417097661484,
+      "required_pages_per_second_upper_estimate": 0.003959785090286927
     },
     {
       "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": 29162.772309203872,
-      "required_pages_per_second_upper_estimate": 0.0017482636709617674
+      "required_token_positions_per_second": 28518.625646492226,
+      "required_pages_per_second_upper_estimate": 0.0017096480620872084
     },
     {
       "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": 29162.772309203872,
-      "required_pages_per_second_upper_estimate": 0.002267277273561572
+      "required_token_positions_per_second": 28518.625646492226,
+      "required_pages_per_second_upper_estimate": 0.0022171977038374785
     },
     {
       "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": 29162.772309203872,
-      "required_pages_per_second_upper_estimate": 0.3740687306371631
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28518.625646492226,
+      "required_pages_per_second_upper_estimate": 0.36580630887869864
     }
   ],
   "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": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11507090302347933,
-    "producer_p95_seconds": 0.21993749499961268,
-    "gpu_update_seconds": 0.9480580140617905,
-    "producer_batches_per_second": 8.690294190147771,
-    "gpu_batches_per_second": 1.0547877716002558,
-    "producer_headroom": 8.238903051524211,
+    "producer_batch_seconds": 0.08155986052349817,
+    "producer_p95_seconds": 0.23460665300081018,
+    "gpu_update_seconds": 0.9694716829175353,
+    "producer_batches_per_second": 12.260933179402514,
+    "gpu_batches_per_second": 1.0314896428852802,
+    "producer_headroom": 11.8866275235748,
     "sustainable": true,
-    "buffer_seconds": 60.67571289995459,
+    "buffer_seconds": 60.107244340887185,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324501.64176173107,
-    "upload_bytes_per_second": 5456801.964260061,
+    "generation_bytes_per_second": 274378.93126705993,
+    "upload_bytes_per_second": 5599824.325670854,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.63049831046055,
+    "estimated_full_shards_per_hour": 58.87533143528794,
     "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": 0.9146980179993989,
-      "bytes": 2516554,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.1637235170001077,
       "bytes": 10485207,
@@ -562,6 +556,12 @@
       "bytes": 7431767,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.270509606001724,
+      "bytes": 15692090,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -183,3 +183,4 @@
 {"step": 543300, "loss": 0.6209806203842163, "lr": 0.0003, "grad_norm": 0.17879850651620527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79521658934, "tokens_per_second": 153872.41254341637, "data_wait_s": 9.962997864931822e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 543400, "loss": 0.5429908633232117, "lr": 0.0003, "grad_norm": 0.15299706480539343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79536370222, "tokens_per_second": 154808.0225924273, "data_wait_s": 9.876002877717838e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 543500, "loss": 0.6088278293609619, "lr": 0.0003, "grad_norm": 0.16523491964644105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79551005916, "tokens_per_second": 113077.61009661564, "data_wait_s": 8.88299837242812e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 543600, "loss": 0.5791146159172058, "lr": 0.0003, "grad_norm": 0.1897782516420057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79565626298, "tokens_per_second": 152548.50523329302, "data_wait_s": 1.2267002603039145e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6395fd8d2cd6490a0abcfb37ecf9aba24c702d9ba3ae764abf29294cd099136
-size 394344
+oid sha256:ef09c8b09dab48c274649b6f12a65155ff8199444d0be329bf31085ad33f2fd2
+size 396462