THE SOURCE IN CONTEXT
Two saved runs found one exact-string needle at 63 percent depth in random filler, with 1034753 and 1035020 prompt tokens. Each completion is seven tokens, with zero cached prompt tokens. These are two different build/setup observations, not repeated measurements of one unchanged configuration or evidence of general long-document reasoning.
What does success mean here?
Two saved runs found one exact-string needle at 63 percent depth in random filler, with 1034753 and 1035020 prompt tokens. Each completion is seven tokens, with zero cached prompt tokens. These are two different build/setup observations, not repeated measurements of one unchanged configuration or evidence of general long-document reasoning.
Source: Elevatormusic · pinned document Elevatormusic · pinned document
How much latency and headroom were observed?
The first token took 830.4 seconds in the earlier run and 817.7 seconds in the final setup. During the final fill, minimum host MemAvailable was 1779 MiB on the head and 2542 MiB on the worker. The source describes engine swap policy, host reclaim settings, Engram page dropping and earlyoom thresholds. This is a tuned narrow-margin observation, not a safe default capacity.
Source: Elevatormusic · pinned document
What is still unknown?
Exact executed image digests and request sampling values are absent from the selected raw records. The author also reports slower prefill states and final-image runs below the best earlier result. One depth, one string and a seven-token output do not validate multiple needles, long generation, multimodal context, concurrent near-million-token requests or general answer correctness.
Source: Elevatormusic · pinned document Elevatormusic · pinned document
What do the saved usage counts establish?
The two saved JSON records retain the actual prompt count, seven completion tokens, zero cached prompt tokens and zero reported reasoning tokens. These values establish the size and outcome of the performed requests. Zero reasoning tokens does not identify an unlogged request thinking flag, and a configured context limit is not an additional completed request.
Source: Elevatormusic · pinned document
Reported figures and their conditions
Selected facts reported by the original author. Each row has its own configuration and measurement boundary. Run AI Here has not reproduced these tests.
| Configuration / workload | Metric | Reported value | Evidence |
|---|---|---|---|
| fd2365042f56 near-1M first token | ttft | 830.4 ssingle prompt, no variance estimate | Original row ↗ |
| 92aa710b533d near-1M first token | ttft | 817.7 ssingle prompt, no variance estimate | Original row ↗ |
| near-1M final head minimum MemAvailable | minimum memavailable | 1779 MiBminimum observed in this one fill, not a safe capacity guarantee | Original row ↗ |
| near-1M final worker minimum MemAvailable | minimum memavailable | 2542 MiBminimum observed in this one fill, not a safe capacity guarantee | Original row ↗ |
fd2365042f56 near-1M first token · ttft · conditions
| model | DeepSeek-V4.1-Flash-1017 |
|---|---|
| model_artifact | LibertAIDAI/DeepSeek-V4.1-Flash-1017-EXL3-2.9bpw |
| model_revision | afc7aa0aab98ee735c0f54e0a122019af1ac08f1 |
| Engram_revision | 1bae0e1088993b9b82de08f59de5be5a34e42170 |
| weight_precision | EXL3 2.9bpw |
| system | two DGX Spark GB10, TP2 |
| node_count | 2 |
| engine | TensorFold Python 0.6.6 |
| runtime_execution_revision | Not specified |
| executed_image_digest | Not specified |
| context_limit_tokens | 1048576 |
| concurrency | 1 |
| needle_depth_fraction | 0.63 |
| workload | one exact-string needle in random filler lines |
| temperature | Not specified |
| top_p | Not specified |
| thinking | Not specified |
| repetitions | 1 |
| general_quality_evaluation | False |
| multi_request_long_context_validated | False |
| runtime_revision_prefix | fd2365042f56 |
| input_tokens | 1034753 |
| output_tokens | 7 |
| cached_tokens | 0 |
| cache_state | zero prefix-cache hit tokens; not a cold OS page-cache guarantee |
| experiment_date | 2026-10-18T01:51:54Z |
| needle_found | True |
| reasoning_tokens | 0 |
| total_seconds | 835.8 |
92aa710b533d near-1M first token · ttft · conditions
| model | DeepSeek-V4.1-Flash-1017 |
|---|---|
| model_artifact | LibertAIDAI/DeepSeek-V4.1-Flash-1017-EXL3-2.9bpw |
| model_revision | afc7aa0aab98ee735c0f54e0a122019af1ac08f1 |
| Engram_revision | 1bae0e1088993b9b82de08f59de5be5a34e42170 |
| weight_precision | EXL3 2.9bpw |
| system | two DGX Spark GB10, TP2 |
| node_count | 2 |
| engine | TensorFold Python 0.6.6 |
| runtime_execution_revision | Not specified |
| executed_image_digest | Not specified |
| context_limit_tokens | 1048576 |
| concurrency | 1 |
| needle_depth_fraction | 0.63 |
| workload | one exact-string needle in random filler lines |
| temperature | Not specified |
| top_p | Not specified |
| thinking | Not specified |
| repetitions | 1 |
| general_quality_evaluation | False |
| multi_request_long_context_validated | False |
| runtime_revision_prefix | 92aa710b533d |
| input_tokens | 1035020 |
| output_tokens | 7 |
| cached_tokens | 0 |
| cache_state | zero prefix-cache hit tokens; not a cold OS page-cache guarantee |
| experiment_date | 2026-10-18T04:47:07Z |
| needle_found | True |
| reasoning_tokens | 0 |
| total_seconds | 825.9 |
near-1M final head minimum MemAvailable · minimum memavailable · conditions
| model | DeepSeek-V4.1-Flash-1017 |
|---|---|
| model_artifact | LibertAIDAI/DeepSeek-V4.1-Flash-1017-EXL3-2.9bpw |
| model_revision | afc7aa0aab98ee735c0f54e0a122019af1ac08f1 |
| Engram_revision | 1bae0e1088993b9b82de08f59de5be5a34e42170 |
| weight_precision | EXL3 2.9bpw |
| system | two DGX Spark GB10, TP2 |
| node_count | 2 |
| engine | TensorFold Python 0.6.6 |
| runtime_execution_revision | Not specified |
| executed_image_digest | Not specified |
| context_limit_tokens | 1048576 |
| concurrency | 1 |
| needle_depth_fraction | 0.63 |
| workload | one exact-string needle in random filler lines |
| temperature | Not specified |
| top_p | Not specified |
| thinking | Not specified |
| repetitions | 1 |
| general_quality_evaluation | False |
| multi_request_long_context_validated | False |
| runtime_revision_prefix | 92aa710b533d |
| input_tokens | 1035020 |
| output_tokens | 7 |
| cache_state | same final needle; cached_tokens 0 |
| experiment_date | 2026-10-18 |
| memory_kind | minimum_host_MemAvailable_during_prefill |
| node | head |
| earlyoom_limit_MiB_at_measurement | 1536 |
| earlyoom_later_limit_MiB | 1024 |
| OS_memory_controls_changed | True |
| ordinary_host_baseline_inferred | False |
near-1M final worker minimum MemAvailable · minimum memavailable · conditions
| model | DeepSeek-V4.1-Flash-1017 |
|---|---|
| model_artifact | LibertAIDAI/DeepSeek-V4.1-Flash-1017-EXL3-2.9bpw |
| model_revision | afc7aa0aab98ee735c0f54e0a122019af1ac08f1 |
| Engram_revision | 1bae0e1088993b9b82de08f59de5be5a34e42170 |
| weight_precision | EXL3 2.9bpw |
| system | two DGX Spark GB10, TP2 |
| node_count | 2 |
| engine | TensorFold Python 0.6.6 |
| runtime_execution_revision | Not specified |
| executed_image_digest | Not specified |
| context_limit_tokens | 1048576 |
| concurrency | 1 |
| needle_depth_fraction | 0.63 |
| workload | one exact-string needle in random filler lines |
| temperature | Not specified |
| top_p | Not specified |
| thinking | Not specified |
| repetitions | 1 |
| general_quality_evaluation | False |
| multi_request_long_context_validated | False |
| runtime_revision_prefix | 92aa710b533d |
| input_tokens | 1035020 |
| output_tokens | 7 |
| cache_state | same final needle; cached_tokens 0 |
| experiment_date | 2026-10-18 |
| memory_kind | minimum_host_MemAvailable_during_prefill |
| node | worker |
| earlyoom_limit_MiB_at_measurement | 1536 |
| earlyoom_later_limit_MiB | 1024 |
| OS_memory_controls_changed | True |
| ordinary_host_baseline_inferred | False |
Original source-grounded editorial synthesis with local Spark Qwen assistance and Codex review; no site model benchmark.